program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] { func main(tensor melspectrogram_features) { string var_90_pad_type_0 = const()[name = string("op_90_pad_type_0"), val = string("custom")]; tensor var_90_pad_0 = const()[name = string("op_90_pad_0"), val = tensor([0, 0, 1, 1])]; tensor var_90_strides_0 = const()[name = string("op_90_strides_0"), val = tensor([1, 1])]; tensor var_90_dilations_0 = const()[name = string("op_90_dilations_0"), val = tensor([1, 1])]; int32 var_90_groups_0 = const()[name = string("op_90_groups_0"), val = int32(1)]; tensor var_65_to_fp16 = const()[name = string("op_65_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor var_71_to_fp16 = const()[name = string("op_71_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368768)))]; tensor var_90_cast_fp16 = conv(bias = var_71_to_fp16, dilations = var_90_dilations_0, groups = var_90_groups_0, pad = var_90_pad_0, pad_type = var_90_pad_type_0, strides = var_90_strides_0, weight = var_65_to_fp16, x = melspectrogram_features)[name = string("op_90_cast_fp16")]; string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_90_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; string var_130_pad_type_0 = const()[name = string("op_130_pad_type_0"), val = string("custom")]; tensor var_130_pad_0 = const()[name = string("op_130_pad_0"), val = tensor([0, 0, 1, 1])]; tensor var_130_strides_0 = const()[name = string("op_130_strides_0"), val = tensor([2, 2])]; tensor var_130_dilations_0 = const()[name = string("op_130_dilations_0"), val = tensor([1, 1])]; int32 var_130_groups_0 = const()[name = string("op_130_groups_0"), val = int32(1)]; tensor var_105_to_fp16 = const()[name = string("op_105_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370368)))]; tensor var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3909376)))]; tensor var_130_cast_fp16 = conv(bias = var_111_to_fp16, dilations = var_130_dilations_0, groups = var_130_groups_0, pad = var_130_pad_0, pad_type = var_130_pad_type_0, strides = var_130_strides_0, weight = var_105_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_130_cast_fp16")]; string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_130_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; tensor var_148_to_fp16 = const()[name = string("op_148_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3910976)))]; tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_148_to_fp16)[name = string("inputs_1_cast_fp16")]; int32 var_158 = const()[name = string("op_158"), val = int32(3)]; int32 var_175 = const()[name = string("op_175"), val = int32(1)]; tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; fp16 var_192_to_fp16 = const()[name = string("op_192_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_192_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6215040)))]; tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6216640)))]; tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6218240)))]; tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6219840)))]; fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6221440)))]; tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7401152)))]; tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7402752)))]; tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8582464)))]; tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9762176)))]; tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; tensor var_234_begin_0 = const()[name = string("op_234_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_234_end_0 = const()[name = string("op_234_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_234_end_mask_0 = const()[name = string("op_234_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = query_1_cast_fp16)[name = string("op_234_cast_fp16")]; tensor var_238_begin_0 = const()[name = string("op_238_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_238_end_0 = const()[name = string("op_238_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_238_end_mask_0 = const()[name = string("op_238_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = query_1_cast_fp16)[name = string("op_238_cast_fp16")]; tensor var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = query_1_cast_fp16)[name = string("op_242_cast_fp16")]; tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = query_1_cast_fp16)[name = string("op_246_cast_fp16")]; tensor var_250_begin_0 = const()[name = string("op_250_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_250_end_0 = const()[name = string("op_250_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_250_end_mask_0 = const()[name = string("op_250_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = query_1_cast_fp16)[name = string("op_250_cast_fp16")]; tensor var_254_begin_0 = const()[name = string("op_254_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_254_end_0 = const()[name = string("op_254_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_254_end_mask_0 = const()[name = string("op_254_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_254_cast_fp16 = slice_by_index(begin = var_254_begin_0, end = var_254_end_0, end_mask = var_254_end_mask_0, x = query_1_cast_fp16)[name = string("op_254_cast_fp16")]; tensor var_258_begin_0 = const()[name = string("op_258_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_258_end_0 = const()[name = string("op_258_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_258_end_mask_0 = const()[name = string("op_258_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_258_cast_fp16 = slice_by_index(begin = var_258_begin_0, end = var_258_end_0, end_mask = var_258_end_mask_0, x = query_1_cast_fp16)[name = string("op_258_cast_fp16")]; tensor var_262_begin_0 = const()[name = string("op_262_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_262_end_0 = const()[name = string("op_262_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_262_end_mask_0 = const()[name = string("op_262_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = query_1_cast_fp16)[name = string("op_262_cast_fp16")]; tensor var_266_begin_0 = const()[name = string("op_266_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_266_end_0 = const()[name = string("op_266_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_266_end_mask_0 = const()[name = string("op_266_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = query_1_cast_fp16)[name = string("op_266_cast_fp16")]; tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = query_1_cast_fp16)[name = string("op_270_cast_fp16")]; tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = query_1_cast_fp16)[name = string("op_274_cast_fp16")]; tensor var_283_begin_0 = const()[name = string("op_283_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_283_end_0 = const()[name = string("op_283_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_283_end_mask_0 = const()[name = string("op_283_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_283_cast_fp16 = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_230_cast_fp16)[name = string("op_283_cast_fp16")]; tensor var_290_begin_0 = const()[name = string("op_290_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_290_end_0 = const()[name = string("op_290_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_290_end_mask_0 = const()[name = string("op_290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_230_cast_fp16)[name = string("op_290_cast_fp16")]; tensor var_297_begin_0 = const()[name = string("op_297_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_297_end_0 = const()[name = string("op_297_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_297_end_mask_0 = const()[name = string("op_297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_297_cast_fp16 = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_230_cast_fp16)[name = string("op_297_cast_fp16")]; tensor var_304_begin_0 = const()[name = string("op_304_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_304_end_0 = const()[name = string("op_304_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_304_end_mask_0 = const()[name = string("op_304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_304_cast_fp16 = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_230_cast_fp16)[name = string("op_304_cast_fp16")]; tensor var_311_begin_0 = const()[name = string("op_311_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_311_end_0 = const()[name = string("op_311_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_311_end_mask_0 = const()[name = string("op_311_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_234_cast_fp16)[name = string("op_311_cast_fp16")]; tensor var_318_begin_0 = const()[name = string("op_318_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_318_end_0 = const()[name = string("op_318_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_318_end_mask_0 = const()[name = string("op_318_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_234_cast_fp16)[name = string("op_318_cast_fp16")]; tensor var_325_begin_0 = const()[name = string("op_325_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_325_end_0 = const()[name = string("op_325_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_325_end_mask_0 = const()[name = string("op_325_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_234_cast_fp16)[name = string("op_325_cast_fp16")]; tensor var_332_begin_0 = const()[name = string("op_332_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_332_end_0 = const()[name = string("op_332_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_332_end_mask_0 = const()[name = string("op_332_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = var_234_cast_fp16)[name = string("op_332_cast_fp16")]; tensor var_339_begin_0 = const()[name = string("op_339_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_339_end_0 = const()[name = string("op_339_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_339_end_mask_0 = const()[name = string("op_339_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_339_cast_fp16 = slice_by_index(begin = var_339_begin_0, end = var_339_end_0, end_mask = var_339_end_mask_0, x = var_238_cast_fp16)[name = string("op_339_cast_fp16")]; tensor var_346_begin_0 = const()[name = string("op_346_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_346_end_0 = const()[name = string("op_346_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_346_end_mask_0 = const()[name = string("op_346_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_346_cast_fp16 = slice_by_index(begin = var_346_begin_0, end = var_346_end_0, end_mask = var_346_end_mask_0, x = var_238_cast_fp16)[name = string("op_346_cast_fp16")]; tensor var_353_begin_0 = const()[name = string("op_353_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_353_end_0 = const()[name = string("op_353_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_353_end_mask_0 = const()[name = string("op_353_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_353_cast_fp16 = slice_by_index(begin = var_353_begin_0, end = var_353_end_0, end_mask = var_353_end_mask_0, x = var_238_cast_fp16)[name = string("op_353_cast_fp16")]; tensor var_360_begin_0 = const()[name = string("op_360_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_360_end_0 = const()[name = string("op_360_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_360_end_mask_0 = const()[name = string("op_360_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_360_cast_fp16 = slice_by_index(begin = var_360_begin_0, end = var_360_end_0, end_mask = var_360_end_mask_0, x = var_238_cast_fp16)[name = string("op_360_cast_fp16")]; tensor var_367_begin_0 = const()[name = string("op_367_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_367_end_0 = const()[name = string("op_367_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_367_end_mask_0 = const()[name = string("op_367_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_367_cast_fp16 = slice_by_index(begin = var_367_begin_0, end = var_367_end_0, end_mask = var_367_end_mask_0, x = var_242_cast_fp16)[name = string("op_367_cast_fp16")]; tensor var_374_begin_0 = const()[name = string("op_374_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_374_end_0 = const()[name = string("op_374_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_374_end_mask_0 = const()[name = string("op_374_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_374_cast_fp16 = slice_by_index(begin = var_374_begin_0, end = var_374_end_0, end_mask = var_374_end_mask_0, x = var_242_cast_fp16)[name = string("op_374_cast_fp16")]; tensor var_381_begin_0 = const()[name = string("op_381_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_381_end_0 = const()[name = string("op_381_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_381_end_mask_0 = const()[name = string("op_381_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_381_cast_fp16 = slice_by_index(begin = var_381_begin_0, end = var_381_end_0, end_mask = var_381_end_mask_0, x = var_242_cast_fp16)[name = string("op_381_cast_fp16")]; tensor var_388_begin_0 = const()[name = string("op_388_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_388_end_0 = const()[name = string("op_388_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_388_end_mask_0 = const()[name = string("op_388_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_388_cast_fp16 = slice_by_index(begin = var_388_begin_0, end = var_388_end_0, end_mask = var_388_end_mask_0, x = var_242_cast_fp16)[name = string("op_388_cast_fp16")]; tensor var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = var_246_cast_fp16)[name = string("op_395_cast_fp16")]; tensor var_402_begin_0 = const()[name = string("op_402_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_402_end_0 = const()[name = string("op_402_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_402_end_mask_0 = const()[name = string("op_402_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_402_cast_fp16 = slice_by_index(begin = var_402_begin_0, end = var_402_end_0, end_mask = var_402_end_mask_0, x = var_246_cast_fp16)[name = string("op_402_cast_fp16")]; tensor var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = var_246_cast_fp16)[name = string("op_409_cast_fp16")]; tensor var_416_begin_0 = const()[name = string("op_416_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_416_end_0 = const()[name = string("op_416_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_416_end_mask_0 = const()[name = string("op_416_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_416_cast_fp16 = slice_by_index(begin = var_416_begin_0, end = var_416_end_0, end_mask = var_416_end_mask_0, x = var_246_cast_fp16)[name = string("op_416_cast_fp16")]; tensor var_423_begin_0 = const()[name = string("op_423_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_423_end_0 = const()[name = string("op_423_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_423_end_mask_0 = const()[name = string("op_423_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_423_cast_fp16 = slice_by_index(begin = var_423_begin_0, end = var_423_end_0, end_mask = var_423_end_mask_0, x = var_250_cast_fp16)[name = string("op_423_cast_fp16")]; tensor var_430_begin_0 = const()[name = string("op_430_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_430_end_0 = const()[name = string("op_430_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_430_end_mask_0 = const()[name = string("op_430_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = var_250_cast_fp16)[name = string("op_430_cast_fp16")]; tensor var_437_begin_0 = const()[name = string("op_437_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_437_end_0 = const()[name = string("op_437_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_437_end_mask_0 = const()[name = string("op_437_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_437_cast_fp16 = slice_by_index(begin = var_437_begin_0, end = var_437_end_0, end_mask = var_437_end_mask_0, x = var_250_cast_fp16)[name = string("op_437_cast_fp16")]; tensor var_444_begin_0 = const()[name = string("op_444_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_444_end_0 = const()[name = string("op_444_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_444_end_mask_0 = const()[name = string("op_444_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_444_cast_fp16 = slice_by_index(begin = var_444_begin_0, end = var_444_end_0, end_mask = var_444_end_mask_0, x = var_250_cast_fp16)[name = string("op_444_cast_fp16")]; tensor var_451_begin_0 = const()[name = string("op_451_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_451_end_0 = const()[name = string("op_451_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_451_end_mask_0 = const()[name = string("op_451_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_451_cast_fp16 = slice_by_index(begin = var_451_begin_0, end = var_451_end_0, end_mask = var_451_end_mask_0, x = var_254_cast_fp16)[name = string("op_451_cast_fp16")]; tensor var_458_begin_0 = const()[name = string("op_458_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_458_end_0 = const()[name = string("op_458_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_458_end_mask_0 = const()[name = string("op_458_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_458_cast_fp16 = slice_by_index(begin = var_458_begin_0, end = var_458_end_0, end_mask = var_458_end_mask_0, x = var_254_cast_fp16)[name = string("op_458_cast_fp16")]; tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = var_254_cast_fp16)[name = string("op_465_cast_fp16")]; tensor var_472_begin_0 = const()[name = string("op_472_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_472_end_0 = const()[name = string("op_472_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_472_end_mask_0 = const()[name = string("op_472_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_472_cast_fp16 = slice_by_index(begin = var_472_begin_0, end = var_472_end_0, end_mask = var_472_end_mask_0, x = var_254_cast_fp16)[name = string("op_472_cast_fp16")]; tensor var_479_begin_0 = const()[name = string("op_479_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_479_end_0 = const()[name = string("op_479_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_479_end_mask_0 = const()[name = string("op_479_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_479_cast_fp16 = slice_by_index(begin = var_479_begin_0, end = var_479_end_0, end_mask = var_479_end_mask_0, x = var_258_cast_fp16)[name = string("op_479_cast_fp16")]; tensor var_486_begin_0 = const()[name = string("op_486_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_486_end_0 = const()[name = string("op_486_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_486_end_mask_0 = const()[name = string("op_486_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_486_cast_fp16 = slice_by_index(begin = var_486_begin_0, end = var_486_end_0, end_mask = var_486_end_mask_0, x = var_258_cast_fp16)[name = string("op_486_cast_fp16")]; tensor var_493_begin_0 = const()[name = string("op_493_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_493_end_0 = const()[name = string("op_493_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_493_end_mask_0 = const()[name = string("op_493_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = var_258_cast_fp16)[name = string("op_493_cast_fp16")]; tensor var_500_begin_0 = const()[name = string("op_500_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_500_end_0 = const()[name = string("op_500_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_500_end_mask_0 = const()[name = string("op_500_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_500_cast_fp16 = slice_by_index(begin = var_500_begin_0, end = var_500_end_0, end_mask = var_500_end_mask_0, x = var_258_cast_fp16)[name = string("op_500_cast_fp16")]; tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = var_262_cast_fp16)[name = string("op_507_cast_fp16")]; tensor var_514_begin_0 = const()[name = string("op_514_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_514_end_0 = const()[name = string("op_514_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_514_end_mask_0 = const()[name = string("op_514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_514_cast_fp16 = slice_by_index(begin = var_514_begin_0, end = var_514_end_0, end_mask = var_514_end_mask_0, x = var_262_cast_fp16)[name = string("op_514_cast_fp16")]; tensor var_521_begin_0 = const()[name = string("op_521_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_521_end_0 = const()[name = string("op_521_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_521_end_mask_0 = const()[name = string("op_521_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_521_cast_fp16 = slice_by_index(begin = var_521_begin_0, end = var_521_end_0, end_mask = var_521_end_mask_0, x = var_262_cast_fp16)[name = string("op_521_cast_fp16")]; tensor var_528_begin_0 = const()[name = string("op_528_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_528_end_0 = const()[name = string("op_528_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_528_end_mask_0 = const()[name = string("op_528_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_528_cast_fp16 = slice_by_index(begin = var_528_begin_0, end = var_528_end_0, end_mask = var_528_end_mask_0, x = var_262_cast_fp16)[name = string("op_528_cast_fp16")]; tensor var_535_begin_0 = const()[name = string("op_535_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_535_end_0 = const()[name = string("op_535_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_535_end_mask_0 = const()[name = string("op_535_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_535_cast_fp16 = slice_by_index(begin = var_535_begin_0, end = var_535_end_0, end_mask = var_535_end_mask_0, x = var_266_cast_fp16)[name = string("op_535_cast_fp16")]; tensor var_542_begin_0 = const()[name = string("op_542_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_542_end_0 = const()[name = string("op_542_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_542_end_mask_0 = const()[name = string("op_542_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_542_cast_fp16 = slice_by_index(begin = var_542_begin_0, end = var_542_end_0, end_mask = var_542_end_mask_0, x = var_266_cast_fp16)[name = string("op_542_cast_fp16")]; tensor var_549_begin_0 = const()[name = string("op_549_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_549_end_0 = const()[name = string("op_549_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_549_end_mask_0 = const()[name = string("op_549_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_549_cast_fp16 = slice_by_index(begin = var_549_begin_0, end = var_549_end_0, end_mask = var_549_end_mask_0, x = var_266_cast_fp16)[name = string("op_549_cast_fp16")]; tensor var_556_begin_0 = const()[name = string("op_556_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_556_end_0 = const()[name = string("op_556_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_556_end_mask_0 = const()[name = string("op_556_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_556_cast_fp16 = slice_by_index(begin = var_556_begin_0, end = var_556_end_0, end_mask = var_556_end_mask_0, x = var_266_cast_fp16)[name = string("op_556_cast_fp16")]; tensor var_563_begin_0 = const()[name = string("op_563_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_563_end_0 = const()[name = string("op_563_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_563_end_mask_0 = const()[name = string("op_563_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_563_cast_fp16 = slice_by_index(begin = var_563_begin_0, end = var_563_end_0, end_mask = var_563_end_mask_0, x = var_270_cast_fp16)[name = string("op_563_cast_fp16")]; tensor var_570_begin_0 = const()[name = string("op_570_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_570_end_0 = const()[name = string("op_570_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_570_end_mask_0 = const()[name = string("op_570_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_570_cast_fp16 = slice_by_index(begin = var_570_begin_0, end = var_570_end_0, end_mask = var_570_end_mask_0, x = var_270_cast_fp16)[name = string("op_570_cast_fp16")]; tensor var_577_begin_0 = const()[name = string("op_577_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_577_end_0 = const()[name = string("op_577_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_577_end_mask_0 = const()[name = string("op_577_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_577_cast_fp16 = slice_by_index(begin = var_577_begin_0, end = var_577_end_0, end_mask = var_577_end_mask_0, x = var_270_cast_fp16)[name = string("op_577_cast_fp16")]; tensor var_584_begin_0 = const()[name = string("op_584_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_584_end_0 = const()[name = string("op_584_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_584_end_mask_0 = const()[name = string("op_584_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_584_cast_fp16 = slice_by_index(begin = var_584_begin_0, end = var_584_end_0, end_mask = var_584_end_mask_0, x = var_270_cast_fp16)[name = string("op_584_cast_fp16")]; tensor var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = var_274_cast_fp16)[name = string("op_591_cast_fp16")]; tensor var_598_begin_0 = const()[name = string("op_598_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_598_end_0 = const()[name = string("op_598_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_598_end_mask_0 = const()[name = string("op_598_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_598_cast_fp16 = slice_by_index(begin = var_598_begin_0, end = var_598_end_0, end_mask = var_598_end_mask_0, x = var_274_cast_fp16)[name = string("op_598_cast_fp16")]; tensor var_605_begin_0 = const()[name = string("op_605_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_605_end_0 = const()[name = string("op_605_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_605_end_mask_0 = const()[name = string("op_605_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_605_cast_fp16 = slice_by_index(begin = var_605_begin_0, end = var_605_end_0, end_mask = var_605_end_mask_0, x = var_274_cast_fp16)[name = string("op_605_cast_fp16")]; tensor var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, x = var_274_cast_fp16)[name = string("op_612_cast_fp16")]; tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_617_begin_0 = const()[name = string("op_617_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_617_end_0 = const()[name = string("op_617_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_617_end_mask_0 = const()[name = string("op_617_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_11")]; tensor var_617_cast_fp16 = slice_by_index(begin = var_617_begin_0, end = var_617_end_0, end_mask = var_617_end_mask_0, x = k_1_cast_fp16)[name = string("op_617_cast_fp16")]; tensor var_621_begin_0 = const()[name = string("op_621_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_621_end_0 = const()[name = string("op_621_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_621_end_mask_0 = const()[name = string("op_621_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_621_cast_fp16 = slice_by_index(begin = var_621_begin_0, end = var_621_end_0, end_mask = var_621_end_mask_0, x = k_1_cast_fp16)[name = string("op_621_cast_fp16")]; tensor var_625_begin_0 = const()[name = string("op_625_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_625_end_0 = const()[name = string("op_625_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_625_end_mask_0 = const()[name = string("op_625_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_625_cast_fp16 = slice_by_index(begin = var_625_begin_0, end = var_625_end_0, end_mask = var_625_end_mask_0, x = k_1_cast_fp16)[name = string("op_625_cast_fp16")]; tensor var_629_begin_0 = const()[name = string("op_629_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_629_end_0 = const()[name = string("op_629_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_629_end_mask_0 = const()[name = string("op_629_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_629_cast_fp16 = slice_by_index(begin = var_629_begin_0, end = var_629_end_0, end_mask = var_629_end_mask_0, x = k_1_cast_fp16)[name = string("op_629_cast_fp16")]; tensor var_633_begin_0 = const()[name = string("op_633_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_633_end_0 = const()[name = string("op_633_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_633_end_mask_0 = const()[name = string("op_633_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_633_cast_fp16 = slice_by_index(begin = var_633_begin_0, end = var_633_end_0, end_mask = var_633_end_mask_0, x = k_1_cast_fp16)[name = string("op_633_cast_fp16")]; tensor var_637_begin_0 = const()[name = string("op_637_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_637_end_0 = const()[name = string("op_637_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_637_end_mask_0 = const()[name = string("op_637_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, x = k_1_cast_fp16)[name = string("op_637_cast_fp16")]; tensor var_641_begin_0 = const()[name = string("op_641_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_641_end_0 = const()[name = string("op_641_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_641_end_mask_0 = const()[name = string("op_641_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_641_cast_fp16 = slice_by_index(begin = var_641_begin_0, end = var_641_end_0, end_mask = var_641_end_mask_0, x = k_1_cast_fp16)[name = string("op_641_cast_fp16")]; tensor var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = k_1_cast_fp16)[name = string("op_645_cast_fp16")]; tensor var_649_begin_0 = const()[name = string("op_649_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_649_end_0 = const()[name = string("op_649_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_649_end_mask_0 = const()[name = string("op_649_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_649_cast_fp16 = slice_by_index(begin = var_649_begin_0, end = var_649_end_0, end_mask = var_649_end_mask_0, x = k_1_cast_fp16)[name = string("op_649_cast_fp16")]; tensor var_653_begin_0 = const()[name = string("op_653_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_653_end_0 = const()[name = string("op_653_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_653_end_mask_0 = const()[name = string("op_653_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_653_cast_fp16 = slice_by_index(begin = var_653_begin_0, end = var_653_end_0, end_mask = var_653_end_mask_0, x = k_1_cast_fp16)[name = string("op_653_cast_fp16")]; tensor var_657_begin_0 = const()[name = string("op_657_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_657_end_0 = const()[name = string("op_657_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_657_end_mask_0 = const()[name = string("op_657_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_657_cast_fp16 = slice_by_index(begin = var_657_begin_0, end = var_657_end_0, end_mask = var_657_end_mask_0, x = k_1_cast_fp16)[name = string("op_657_cast_fp16")]; tensor var_661_begin_0 = const()[name = string("op_661_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_661_end_0 = const()[name = string("op_661_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_661_end_mask_0 = const()[name = string("op_661_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_661_cast_fp16 = slice_by_index(begin = var_661_begin_0, end = var_661_end_0, end_mask = var_661_end_mask_0, x = k_1_cast_fp16)[name = string("op_661_cast_fp16")]; tensor var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = value_1_cast_fp16)[name = string("op_663_cast_fp16")]; tensor var_667_begin_0 = const()[name = string("op_667_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_667_end_0 = const()[name = string("op_667_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_667_end_mask_0 = const()[name = string("op_667_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_667_cast_fp16 = slice_by_index(begin = var_667_begin_0, end = var_667_end_0, end_mask = var_667_end_mask_0, x = value_1_cast_fp16)[name = string("op_667_cast_fp16")]; tensor var_671_begin_0 = const()[name = string("op_671_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_671_end_0 = const()[name = string("op_671_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_671_end_mask_0 = const()[name = string("op_671_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_671_cast_fp16 = slice_by_index(begin = var_671_begin_0, end = var_671_end_0, end_mask = var_671_end_mask_0, x = value_1_cast_fp16)[name = string("op_671_cast_fp16")]; tensor var_675_begin_0 = const()[name = string("op_675_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_675_end_0 = const()[name = string("op_675_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_675_end_mask_0 = const()[name = string("op_675_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_675_cast_fp16 = slice_by_index(begin = var_675_begin_0, end = var_675_end_0, end_mask = var_675_end_mask_0, x = value_1_cast_fp16)[name = string("op_675_cast_fp16")]; tensor var_679_begin_0 = const()[name = string("op_679_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_679_end_0 = const()[name = string("op_679_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_679_end_mask_0 = const()[name = string("op_679_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, x = value_1_cast_fp16)[name = string("op_679_cast_fp16")]; tensor var_683_begin_0 = const()[name = string("op_683_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_683_end_0 = const()[name = string("op_683_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_683_end_mask_0 = const()[name = string("op_683_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, x = value_1_cast_fp16)[name = string("op_683_cast_fp16")]; tensor var_687_begin_0 = const()[name = string("op_687_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_687_end_0 = const()[name = string("op_687_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_687_end_mask_0 = const()[name = string("op_687_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_687_cast_fp16 = slice_by_index(begin = var_687_begin_0, end = var_687_end_0, end_mask = var_687_end_mask_0, x = value_1_cast_fp16)[name = string("op_687_cast_fp16")]; tensor var_691_begin_0 = const()[name = string("op_691_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_691_end_0 = const()[name = string("op_691_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_691_end_mask_0 = const()[name = string("op_691_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_691_cast_fp16 = slice_by_index(begin = var_691_begin_0, end = var_691_end_0, end_mask = var_691_end_mask_0, x = value_1_cast_fp16)[name = string("op_691_cast_fp16")]; tensor var_695_begin_0 = const()[name = string("op_695_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_695_end_0 = const()[name = string("op_695_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_695_end_mask_0 = const()[name = string("op_695_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_695_cast_fp16 = slice_by_index(begin = var_695_begin_0, end = var_695_end_0, end_mask = var_695_end_mask_0, x = value_1_cast_fp16)[name = string("op_695_cast_fp16")]; tensor var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = value_1_cast_fp16)[name = string("op_699_cast_fp16")]; tensor var_703_begin_0 = const()[name = string("op_703_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_703_end_0 = const()[name = string("op_703_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_703_end_mask_0 = const()[name = string("op_703_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = value_1_cast_fp16)[name = string("op_703_cast_fp16")]; tensor var_707_begin_0 = const()[name = string("op_707_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_707_end_0 = const()[name = string("op_707_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_707_end_mask_0 = const()[name = string("op_707_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = value_1_cast_fp16)[name = string("op_707_cast_fp16")]; string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_617_cast_fp16, var_283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_617_cast_fp16, var_290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_617_cast_fp16, var_297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_617_cast_fp16, var_304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_621_cast_fp16, var_311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_621_cast_fp16, var_318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_621_cast_fp16, var_325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_621_cast_fp16, var_332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_625_cast_fp16, var_339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_625_cast_fp16, var_346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_625_cast_fp16, var_353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_625_cast_fp16, var_360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_629_cast_fp16, var_367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_629_cast_fp16, var_374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_629_cast_fp16, var_381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_629_cast_fp16, var_388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_633_cast_fp16, var_395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_633_cast_fp16, var_402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_633_cast_fp16, var_409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_633_cast_fp16, var_416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_637_cast_fp16, var_423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_637_cast_fp16, var_430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_637_cast_fp16, var_437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_637_cast_fp16, var_444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_641_cast_fp16, var_451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_641_cast_fp16, var_458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_641_cast_fp16, var_465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_641_cast_fp16, var_472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_645_cast_fp16, var_479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_645_cast_fp16, var_486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_645_cast_fp16, var_493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_645_cast_fp16, var_500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_649_cast_fp16, var_507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_649_cast_fp16, var_514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_649_cast_fp16, var_521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_649_cast_fp16, var_528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_653_cast_fp16, var_535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_653_cast_fp16, var_542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_653_cast_fp16, var_549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_653_cast_fp16, var_556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_657_cast_fp16, var_563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_657_cast_fp16, var_570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_657_cast_fp16, var_577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_657_cast_fp16, var_584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_661_cast_fp16, var_591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_661_cast_fp16, var_598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_661_cast_fp16, var_605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_661_cast_fp16, var_612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; fp16 var_806_to_fp16 = const()[name = string("op_806_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_806_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; fp16 var_808_to_fp16 = const()[name = string("op_808_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_808_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; fp16 var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_810_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_812_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; fp16 var_814_to_fp16 = const()[name = string("op_814_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_814_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; fp16 var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_816_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; fp16 var_818_to_fp16 = const()[name = string("op_818_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_818_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; fp16 var_820_to_fp16 = const()[name = string("op_820_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_820_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; fp16 var_822_to_fp16 = const()[name = string("op_822_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_822_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; fp16 var_824_to_fp16 = const()[name = string("op_824_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_824_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; fp16 var_826_to_fp16 = const()[name = string("op_826_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_826_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; fp16 var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_828_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; fp16 var_830_to_fp16 = const()[name = string("op_830_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_830_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; fp16 var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_832_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; fp16 var_834_to_fp16 = const()[name = string("op_834_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_834_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; fp16 var_836_to_fp16 = const()[name = string("op_836_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_836_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; fp16 var_838_to_fp16 = const()[name = string("op_838_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_838_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; fp16 var_840_to_fp16 = const()[name = string("op_840_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_840_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; fp16 var_842_to_fp16 = const()[name = string("op_842_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_842_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; fp16 var_844_to_fp16 = const()[name = string("op_844_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_844_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; fp16 var_846_to_fp16 = const()[name = string("op_846_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_846_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; fp16 var_848_to_fp16 = const()[name = string("op_848_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_848_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; fp16 var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_850_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; fp16 var_852_to_fp16 = const()[name = string("op_852_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_852_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; fp16 var_854_to_fp16 = const()[name = string("op_854_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_854_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; fp16 var_856_to_fp16 = const()[name = string("op_856_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_856_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; fp16 var_858_to_fp16 = const()[name = string("op_858_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_858_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; fp16 var_860_to_fp16 = const()[name = string("op_860_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_860_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; fp16 var_862_to_fp16 = const()[name = string("op_862_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_862_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; fp16 var_864_to_fp16 = const()[name = string("op_864_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_864_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; fp16 var_866_to_fp16 = const()[name = string("op_866_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_866_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; fp16 var_868_to_fp16 = const()[name = string("op_868_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_868_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; fp16 var_870_to_fp16 = const()[name = string("op_870_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_870_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; fp16 var_872_to_fp16 = const()[name = string("op_872_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_872_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_874_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; fp16 var_876_to_fp16 = const()[name = string("op_876_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_876_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_878_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; fp16 var_880_to_fp16 = const()[name = string("op_880_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_880_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; fp16 var_882_to_fp16 = const()[name = string("op_882_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_882_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_884_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; fp16 var_886_to_fp16 = const()[name = string("op_886_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_886_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; fp16 var_888_to_fp16 = const()[name = string("op_888_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_888_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; fp16 var_890_to_fp16 = const()[name = string("op_890_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_890_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; fp16 var_892_to_fp16 = const()[name = string("op_892_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_892_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; fp16 var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_894_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; fp16 var_896_to_fp16 = const()[name = string("op_896_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_896_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; fp16 var_898_to_fp16 = const()[name = string("op_898_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_898_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; fp16 var_900_to_fp16 = const()[name = string("op_900_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_900_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; tensor var_902_cast_fp16 = softmax(axis = var_175, x = aw_chunk_1_cast_fp16)[name = string("op_902_cast_fp16")]; tensor var_903_cast_fp16 = softmax(axis = var_175, x = aw_chunk_3_cast_fp16)[name = string("op_903_cast_fp16")]; tensor var_904_cast_fp16 = softmax(axis = var_175, x = aw_chunk_5_cast_fp16)[name = string("op_904_cast_fp16")]; tensor var_905_cast_fp16 = softmax(axis = var_175, x = aw_chunk_7_cast_fp16)[name = string("op_905_cast_fp16")]; tensor var_906_cast_fp16 = softmax(axis = var_175, x = aw_chunk_9_cast_fp16)[name = string("op_906_cast_fp16")]; tensor var_907_cast_fp16 = softmax(axis = var_175, x = aw_chunk_11_cast_fp16)[name = string("op_907_cast_fp16")]; tensor var_908_cast_fp16 = softmax(axis = var_175, x = aw_chunk_13_cast_fp16)[name = string("op_908_cast_fp16")]; tensor var_909_cast_fp16 = softmax(axis = var_175, x = aw_chunk_15_cast_fp16)[name = string("op_909_cast_fp16")]; tensor var_910_cast_fp16 = softmax(axis = var_175, x = aw_chunk_17_cast_fp16)[name = string("op_910_cast_fp16")]; tensor var_911_cast_fp16 = softmax(axis = var_175, x = aw_chunk_19_cast_fp16)[name = string("op_911_cast_fp16")]; tensor var_912_cast_fp16 = softmax(axis = var_175, x = aw_chunk_21_cast_fp16)[name = string("op_912_cast_fp16")]; tensor var_913_cast_fp16 = softmax(axis = var_175, x = aw_chunk_23_cast_fp16)[name = string("op_913_cast_fp16")]; tensor var_914_cast_fp16 = softmax(axis = var_175, x = aw_chunk_25_cast_fp16)[name = string("op_914_cast_fp16")]; tensor var_915_cast_fp16 = softmax(axis = var_175, x = aw_chunk_27_cast_fp16)[name = string("op_915_cast_fp16")]; tensor var_916_cast_fp16 = softmax(axis = var_175, x = aw_chunk_29_cast_fp16)[name = string("op_916_cast_fp16")]; tensor var_917_cast_fp16 = softmax(axis = var_175, x = aw_chunk_31_cast_fp16)[name = string("op_917_cast_fp16")]; tensor var_918_cast_fp16 = softmax(axis = var_175, x = aw_chunk_33_cast_fp16)[name = string("op_918_cast_fp16")]; tensor var_919_cast_fp16 = softmax(axis = var_175, x = aw_chunk_35_cast_fp16)[name = string("op_919_cast_fp16")]; tensor var_920_cast_fp16 = softmax(axis = var_175, x = aw_chunk_37_cast_fp16)[name = string("op_920_cast_fp16")]; tensor var_921_cast_fp16 = softmax(axis = var_175, x = aw_chunk_39_cast_fp16)[name = string("op_921_cast_fp16")]; tensor var_922_cast_fp16 = softmax(axis = var_175, x = aw_chunk_41_cast_fp16)[name = string("op_922_cast_fp16")]; tensor var_923_cast_fp16 = softmax(axis = var_175, x = aw_chunk_43_cast_fp16)[name = string("op_923_cast_fp16")]; tensor var_924_cast_fp16 = softmax(axis = var_175, x = aw_chunk_45_cast_fp16)[name = string("op_924_cast_fp16")]; tensor var_925_cast_fp16 = softmax(axis = var_175, x = aw_chunk_47_cast_fp16)[name = string("op_925_cast_fp16")]; tensor var_926_cast_fp16 = softmax(axis = var_175, x = aw_chunk_49_cast_fp16)[name = string("op_926_cast_fp16")]; tensor var_927_cast_fp16 = softmax(axis = var_175, x = aw_chunk_51_cast_fp16)[name = string("op_927_cast_fp16")]; tensor var_928_cast_fp16 = softmax(axis = var_175, x = aw_chunk_53_cast_fp16)[name = string("op_928_cast_fp16")]; tensor var_929_cast_fp16 = softmax(axis = var_175, x = aw_chunk_55_cast_fp16)[name = string("op_929_cast_fp16")]; tensor var_930_cast_fp16 = softmax(axis = var_175, x = aw_chunk_57_cast_fp16)[name = string("op_930_cast_fp16")]; tensor var_931_cast_fp16 = softmax(axis = var_175, x = aw_chunk_59_cast_fp16)[name = string("op_931_cast_fp16")]; tensor var_932_cast_fp16 = softmax(axis = var_175, x = aw_chunk_61_cast_fp16)[name = string("op_932_cast_fp16")]; tensor var_933_cast_fp16 = softmax(axis = var_175, x = aw_chunk_63_cast_fp16)[name = string("op_933_cast_fp16")]; tensor var_934_cast_fp16 = softmax(axis = var_175, x = aw_chunk_65_cast_fp16)[name = string("op_934_cast_fp16")]; tensor var_935_cast_fp16 = softmax(axis = var_175, x = aw_chunk_67_cast_fp16)[name = string("op_935_cast_fp16")]; tensor var_936_cast_fp16 = softmax(axis = var_175, x = aw_chunk_69_cast_fp16)[name = string("op_936_cast_fp16")]; tensor var_937_cast_fp16 = softmax(axis = var_175, x = aw_chunk_71_cast_fp16)[name = string("op_937_cast_fp16")]; tensor var_938_cast_fp16 = softmax(axis = var_175, x = aw_chunk_73_cast_fp16)[name = string("op_938_cast_fp16")]; tensor var_939_cast_fp16 = softmax(axis = var_175, x = aw_chunk_75_cast_fp16)[name = string("op_939_cast_fp16")]; tensor var_940_cast_fp16 = softmax(axis = var_175, x = aw_chunk_77_cast_fp16)[name = string("op_940_cast_fp16")]; tensor var_941_cast_fp16 = softmax(axis = var_175, x = aw_chunk_79_cast_fp16)[name = string("op_941_cast_fp16")]; tensor var_942_cast_fp16 = softmax(axis = var_175, x = aw_chunk_81_cast_fp16)[name = string("op_942_cast_fp16")]; tensor var_943_cast_fp16 = softmax(axis = var_175, x = aw_chunk_83_cast_fp16)[name = string("op_943_cast_fp16")]; tensor var_944_cast_fp16 = softmax(axis = var_175, x = aw_chunk_85_cast_fp16)[name = string("op_944_cast_fp16")]; tensor var_945_cast_fp16 = softmax(axis = var_175, x = aw_chunk_87_cast_fp16)[name = string("op_945_cast_fp16")]; tensor var_946_cast_fp16 = softmax(axis = var_175, x = aw_chunk_89_cast_fp16)[name = string("op_946_cast_fp16")]; tensor var_947_cast_fp16 = softmax(axis = var_175, x = aw_chunk_91_cast_fp16)[name = string("op_947_cast_fp16")]; tensor var_948_cast_fp16 = softmax(axis = var_175, x = aw_chunk_93_cast_fp16)[name = string("op_948_cast_fp16")]; tensor var_949_cast_fp16 = softmax(axis = var_175, x = aw_chunk_95_cast_fp16)[name = string("op_949_cast_fp16")]; string var_951_equation_0 = const()[name = string("op_951_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_951_cast_fp16 = einsum(equation = var_951_equation_0, values = (var_663_cast_fp16, var_902_cast_fp16))[name = string("op_951_cast_fp16")]; string var_953_equation_0 = const()[name = string("op_953_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_953_cast_fp16 = einsum(equation = var_953_equation_0, values = (var_663_cast_fp16, var_903_cast_fp16))[name = string("op_953_cast_fp16")]; string var_955_equation_0 = const()[name = string("op_955_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_955_cast_fp16 = einsum(equation = var_955_equation_0, values = (var_663_cast_fp16, var_904_cast_fp16))[name = string("op_955_cast_fp16")]; string var_957_equation_0 = const()[name = string("op_957_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_957_cast_fp16 = einsum(equation = var_957_equation_0, values = (var_663_cast_fp16, var_905_cast_fp16))[name = string("op_957_cast_fp16")]; string var_959_equation_0 = const()[name = string("op_959_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_959_cast_fp16 = einsum(equation = var_959_equation_0, values = (var_667_cast_fp16, var_906_cast_fp16))[name = string("op_959_cast_fp16")]; string var_961_equation_0 = const()[name = string("op_961_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_961_cast_fp16 = einsum(equation = var_961_equation_0, values = (var_667_cast_fp16, var_907_cast_fp16))[name = string("op_961_cast_fp16")]; string var_963_equation_0 = const()[name = string("op_963_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_963_cast_fp16 = einsum(equation = var_963_equation_0, values = (var_667_cast_fp16, var_908_cast_fp16))[name = string("op_963_cast_fp16")]; string var_965_equation_0 = const()[name = string("op_965_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_965_cast_fp16 = einsum(equation = var_965_equation_0, values = (var_667_cast_fp16, var_909_cast_fp16))[name = string("op_965_cast_fp16")]; string var_967_equation_0 = const()[name = string("op_967_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_967_cast_fp16 = einsum(equation = var_967_equation_0, values = (var_671_cast_fp16, var_910_cast_fp16))[name = string("op_967_cast_fp16")]; string var_969_equation_0 = const()[name = string("op_969_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_969_cast_fp16 = einsum(equation = var_969_equation_0, values = (var_671_cast_fp16, var_911_cast_fp16))[name = string("op_969_cast_fp16")]; string var_971_equation_0 = const()[name = string("op_971_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_971_cast_fp16 = einsum(equation = var_971_equation_0, values = (var_671_cast_fp16, var_912_cast_fp16))[name = string("op_971_cast_fp16")]; string var_973_equation_0 = const()[name = string("op_973_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_973_cast_fp16 = einsum(equation = var_973_equation_0, values = (var_671_cast_fp16, var_913_cast_fp16))[name = string("op_973_cast_fp16")]; string var_975_equation_0 = const()[name = string("op_975_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_975_cast_fp16 = einsum(equation = var_975_equation_0, values = (var_675_cast_fp16, var_914_cast_fp16))[name = string("op_975_cast_fp16")]; string var_977_equation_0 = const()[name = string("op_977_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_977_cast_fp16 = einsum(equation = var_977_equation_0, values = (var_675_cast_fp16, var_915_cast_fp16))[name = string("op_977_cast_fp16")]; string var_979_equation_0 = const()[name = string("op_979_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_979_cast_fp16 = einsum(equation = var_979_equation_0, values = (var_675_cast_fp16, var_916_cast_fp16))[name = string("op_979_cast_fp16")]; string var_981_equation_0 = const()[name = string("op_981_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_981_cast_fp16 = einsum(equation = var_981_equation_0, values = (var_675_cast_fp16, var_917_cast_fp16))[name = string("op_981_cast_fp16")]; string var_983_equation_0 = const()[name = string("op_983_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_983_cast_fp16 = einsum(equation = var_983_equation_0, values = (var_679_cast_fp16, var_918_cast_fp16))[name = string("op_983_cast_fp16")]; string var_985_equation_0 = const()[name = string("op_985_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_985_cast_fp16 = einsum(equation = var_985_equation_0, values = (var_679_cast_fp16, var_919_cast_fp16))[name = string("op_985_cast_fp16")]; string var_987_equation_0 = const()[name = string("op_987_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_987_cast_fp16 = einsum(equation = var_987_equation_0, values = (var_679_cast_fp16, var_920_cast_fp16))[name = string("op_987_cast_fp16")]; string var_989_equation_0 = const()[name = string("op_989_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_989_cast_fp16 = einsum(equation = var_989_equation_0, values = (var_679_cast_fp16, var_921_cast_fp16))[name = string("op_989_cast_fp16")]; string var_991_equation_0 = const()[name = string("op_991_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_991_cast_fp16 = einsum(equation = var_991_equation_0, values = (var_683_cast_fp16, var_922_cast_fp16))[name = string("op_991_cast_fp16")]; string var_993_equation_0 = const()[name = string("op_993_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_993_cast_fp16 = einsum(equation = var_993_equation_0, values = (var_683_cast_fp16, var_923_cast_fp16))[name = string("op_993_cast_fp16")]; string var_995_equation_0 = const()[name = string("op_995_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_995_cast_fp16 = einsum(equation = var_995_equation_0, values = (var_683_cast_fp16, var_924_cast_fp16))[name = string("op_995_cast_fp16")]; string var_997_equation_0 = const()[name = string("op_997_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_997_cast_fp16 = einsum(equation = var_997_equation_0, values = (var_683_cast_fp16, var_925_cast_fp16))[name = string("op_997_cast_fp16")]; string var_999_equation_0 = const()[name = string("op_999_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_999_cast_fp16 = einsum(equation = var_999_equation_0, values = (var_687_cast_fp16, var_926_cast_fp16))[name = string("op_999_cast_fp16")]; string var_1001_equation_0 = const()[name = string("op_1001_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1001_cast_fp16 = einsum(equation = var_1001_equation_0, values = (var_687_cast_fp16, var_927_cast_fp16))[name = string("op_1001_cast_fp16")]; string var_1003_equation_0 = const()[name = string("op_1003_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1003_cast_fp16 = einsum(equation = var_1003_equation_0, values = (var_687_cast_fp16, var_928_cast_fp16))[name = string("op_1003_cast_fp16")]; string var_1005_equation_0 = const()[name = string("op_1005_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1005_cast_fp16 = einsum(equation = var_1005_equation_0, values = (var_687_cast_fp16, var_929_cast_fp16))[name = string("op_1005_cast_fp16")]; string var_1007_equation_0 = const()[name = string("op_1007_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1007_cast_fp16 = einsum(equation = var_1007_equation_0, values = (var_691_cast_fp16, var_930_cast_fp16))[name = string("op_1007_cast_fp16")]; string var_1009_equation_0 = const()[name = string("op_1009_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1009_cast_fp16 = einsum(equation = var_1009_equation_0, values = (var_691_cast_fp16, var_931_cast_fp16))[name = string("op_1009_cast_fp16")]; string var_1011_equation_0 = const()[name = string("op_1011_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1011_cast_fp16 = einsum(equation = var_1011_equation_0, values = (var_691_cast_fp16, var_932_cast_fp16))[name = string("op_1011_cast_fp16")]; string var_1013_equation_0 = const()[name = string("op_1013_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1013_cast_fp16 = einsum(equation = var_1013_equation_0, values = (var_691_cast_fp16, var_933_cast_fp16))[name = string("op_1013_cast_fp16")]; string var_1015_equation_0 = const()[name = string("op_1015_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1015_cast_fp16 = einsum(equation = var_1015_equation_0, values = (var_695_cast_fp16, var_934_cast_fp16))[name = string("op_1015_cast_fp16")]; string var_1017_equation_0 = const()[name = string("op_1017_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1017_cast_fp16 = einsum(equation = var_1017_equation_0, values = (var_695_cast_fp16, var_935_cast_fp16))[name = string("op_1017_cast_fp16")]; string var_1019_equation_0 = const()[name = string("op_1019_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1019_cast_fp16 = einsum(equation = var_1019_equation_0, values = (var_695_cast_fp16, var_936_cast_fp16))[name = string("op_1019_cast_fp16")]; string var_1021_equation_0 = const()[name = string("op_1021_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1021_cast_fp16 = einsum(equation = var_1021_equation_0, values = (var_695_cast_fp16, var_937_cast_fp16))[name = string("op_1021_cast_fp16")]; string var_1023_equation_0 = const()[name = string("op_1023_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1023_cast_fp16 = einsum(equation = var_1023_equation_0, values = (var_699_cast_fp16, var_938_cast_fp16))[name = string("op_1023_cast_fp16")]; string var_1025_equation_0 = const()[name = string("op_1025_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1025_cast_fp16 = einsum(equation = var_1025_equation_0, values = (var_699_cast_fp16, var_939_cast_fp16))[name = string("op_1025_cast_fp16")]; string var_1027_equation_0 = const()[name = string("op_1027_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1027_cast_fp16 = einsum(equation = var_1027_equation_0, values = (var_699_cast_fp16, var_940_cast_fp16))[name = string("op_1027_cast_fp16")]; string var_1029_equation_0 = const()[name = string("op_1029_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1029_cast_fp16 = einsum(equation = var_1029_equation_0, values = (var_699_cast_fp16, var_941_cast_fp16))[name = string("op_1029_cast_fp16")]; string var_1031_equation_0 = const()[name = string("op_1031_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1031_cast_fp16 = einsum(equation = var_1031_equation_0, values = (var_703_cast_fp16, var_942_cast_fp16))[name = string("op_1031_cast_fp16")]; string var_1033_equation_0 = const()[name = string("op_1033_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1033_cast_fp16 = einsum(equation = var_1033_equation_0, values = (var_703_cast_fp16, var_943_cast_fp16))[name = string("op_1033_cast_fp16")]; string var_1035_equation_0 = const()[name = string("op_1035_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1035_cast_fp16 = einsum(equation = var_1035_equation_0, values = (var_703_cast_fp16, var_944_cast_fp16))[name = string("op_1035_cast_fp16")]; string var_1037_equation_0 = const()[name = string("op_1037_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1037_cast_fp16 = einsum(equation = var_1037_equation_0, values = (var_703_cast_fp16, var_945_cast_fp16))[name = string("op_1037_cast_fp16")]; string var_1039_equation_0 = const()[name = string("op_1039_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1039_cast_fp16 = einsum(equation = var_1039_equation_0, values = (var_707_cast_fp16, var_946_cast_fp16))[name = string("op_1039_cast_fp16")]; string var_1041_equation_0 = const()[name = string("op_1041_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1041_cast_fp16 = einsum(equation = var_1041_equation_0, values = (var_707_cast_fp16, var_947_cast_fp16))[name = string("op_1041_cast_fp16")]; string var_1043_equation_0 = const()[name = string("op_1043_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1043_cast_fp16 = einsum(equation = var_1043_equation_0, values = (var_707_cast_fp16, var_948_cast_fp16))[name = string("op_1043_cast_fp16")]; string var_1045_equation_0 = const()[name = string("op_1045_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1045_cast_fp16 = einsum(equation = var_1045_equation_0, values = (var_707_cast_fp16, var_949_cast_fp16))[name = string("op_1045_cast_fp16")]; bool var_1047_interleave_0 = const()[name = string("op_1047_interleave_0"), val = bool(false)]; tensor var_1047_cast_fp16 = concat(axis = var_158, interleave = var_1047_interleave_0, values = (var_951_cast_fp16, var_953_cast_fp16, var_955_cast_fp16, var_957_cast_fp16))[name = string("op_1047_cast_fp16")]; bool var_1049_interleave_0 = const()[name = string("op_1049_interleave_0"), val = bool(false)]; tensor var_1049_cast_fp16 = concat(axis = var_158, interleave = var_1049_interleave_0, values = (var_959_cast_fp16, var_961_cast_fp16, var_963_cast_fp16, var_965_cast_fp16))[name = string("op_1049_cast_fp16")]; bool var_1051_interleave_0 = const()[name = string("op_1051_interleave_0"), val = bool(false)]; tensor var_1051_cast_fp16 = concat(axis = var_158, interleave = var_1051_interleave_0, values = (var_967_cast_fp16, var_969_cast_fp16, var_971_cast_fp16, var_973_cast_fp16))[name = string("op_1051_cast_fp16")]; bool var_1053_interleave_0 = const()[name = string("op_1053_interleave_0"), val = bool(false)]; tensor var_1053_cast_fp16 = concat(axis = var_158, interleave = var_1053_interleave_0, values = (var_975_cast_fp16, var_977_cast_fp16, var_979_cast_fp16, var_981_cast_fp16))[name = string("op_1053_cast_fp16")]; bool var_1055_interleave_0 = const()[name = string("op_1055_interleave_0"), val = bool(false)]; tensor var_1055_cast_fp16 = concat(axis = var_158, interleave = var_1055_interleave_0, values = (var_983_cast_fp16, var_985_cast_fp16, var_987_cast_fp16, var_989_cast_fp16))[name = string("op_1055_cast_fp16")]; bool var_1057_interleave_0 = const()[name = string("op_1057_interleave_0"), val = bool(false)]; tensor var_1057_cast_fp16 = concat(axis = var_158, interleave = var_1057_interleave_0, values = (var_991_cast_fp16, var_993_cast_fp16, var_995_cast_fp16, var_997_cast_fp16))[name = string("op_1057_cast_fp16")]; bool var_1059_interleave_0 = const()[name = string("op_1059_interleave_0"), val = bool(false)]; tensor var_1059_cast_fp16 = concat(axis = var_158, interleave = var_1059_interleave_0, values = (var_999_cast_fp16, var_1001_cast_fp16, var_1003_cast_fp16, var_1005_cast_fp16))[name = string("op_1059_cast_fp16")]; bool var_1061_interleave_0 = const()[name = string("op_1061_interleave_0"), val = bool(false)]; tensor var_1061_cast_fp16 = concat(axis = var_158, interleave = var_1061_interleave_0, values = (var_1007_cast_fp16, var_1009_cast_fp16, var_1011_cast_fp16, var_1013_cast_fp16))[name = string("op_1061_cast_fp16")]; bool var_1063_interleave_0 = const()[name = string("op_1063_interleave_0"), val = bool(false)]; tensor var_1063_cast_fp16 = concat(axis = var_158, interleave = var_1063_interleave_0, values = (var_1015_cast_fp16, var_1017_cast_fp16, var_1019_cast_fp16, var_1021_cast_fp16))[name = string("op_1063_cast_fp16")]; bool var_1065_interleave_0 = const()[name = string("op_1065_interleave_0"), val = bool(false)]; tensor var_1065_cast_fp16 = concat(axis = var_158, interleave = var_1065_interleave_0, values = (var_1023_cast_fp16, var_1025_cast_fp16, var_1027_cast_fp16, var_1029_cast_fp16))[name = string("op_1065_cast_fp16")]; bool var_1067_interleave_0 = const()[name = string("op_1067_interleave_0"), val = bool(false)]; tensor var_1067_cast_fp16 = concat(axis = var_158, interleave = var_1067_interleave_0, values = (var_1031_cast_fp16, var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16))[name = string("op_1067_cast_fp16")]; bool var_1069_interleave_0 = const()[name = string("op_1069_interleave_0"), val = bool(false)]; tensor var_1069_cast_fp16 = concat(axis = var_158, interleave = var_1069_interleave_0, values = (var_1039_cast_fp16, var_1041_cast_fp16, var_1043_cast_fp16, var_1045_cast_fp16))[name = string("op_1069_cast_fp16")]; bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; tensor input_1_cast_fp16 = concat(axis = var_175, interleave = input_1_interleave_0, values = (var_1047_cast_fp16, var_1049_cast_fp16, var_1051_cast_fp16, var_1053_cast_fp16, var_1055_cast_fp16, var_1057_cast_fp16, var_1059_cast_fp16, var_1061_cast_fp16, var_1063_cast_fp16, var_1065_cast_fp16, var_1067_cast_fp16, var_1069_cast_fp16))[name = string("input_1_cast_fp16")]; string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9763776)))]; tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10943488)))]; tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; fp16 var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_1088_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10945088)))]; tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10946688)))]; fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10948288)))]; tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15666944)))]; tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15673152)))]; tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20391808)))]; tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; int32 var_1117 = const()[name = string("op_1117"), val = int32(3)]; int32 var_1134 = const()[name = string("op_1134"), val = int32(1)]; tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; fp16 var_1151_to_fp16 = const()[name = string("op_1151_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_1151_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20393408)))]; tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20395008)))]; fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20396608)))]; tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21576320)))]; tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21577920)))]; tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22757632)))]; tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937344)))]; tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; tensor var_1189_begin_0 = const()[name = string("op_1189_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1189_end_0 = const()[name = string("op_1189_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1189_end_mask_0 = const()[name = string("op_1189_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1189_cast_fp16 = slice_by_index(begin = var_1189_begin_0, end = var_1189_end_0, end_mask = var_1189_end_mask_0, x = query_3_cast_fp16)[name = string("op_1189_cast_fp16")]; tensor var_1193_begin_0 = const()[name = string("op_1193_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_1193_end_0 = const()[name = string("op_1193_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_1193_end_mask_0 = const()[name = string("op_1193_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1193_cast_fp16 = slice_by_index(begin = var_1193_begin_0, end = var_1193_end_0, end_mask = var_1193_end_mask_0, x = query_3_cast_fp16)[name = string("op_1193_cast_fp16")]; tensor var_1197_begin_0 = const()[name = string("op_1197_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_1197_end_0 = const()[name = string("op_1197_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_1197_end_mask_0 = const()[name = string("op_1197_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1197_cast_fp16 = slice_by_index(begin = var_1197_begin_0, end = var_1197_end_0, end_mask = var_1197_end_mask_0, x = query_3_cast_fp16)[name = string("op_1197_cast_fp16")]; tensor var_1201_begin_0 = const()[name = string("op_1201_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_1201_end_0 = const()[name = string("op_1201_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_1201_end_mask_0 = const()[name = string("op_1201_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = query_3_cast_fp16)[name = string("op_1201_cast_fp16")]; tensor var_1205_begin_0 = const()[name = string("op_1205_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_1205_end_0 = const()[name = string("op_1205_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_1205_end_mask_0 = const()[name = string("op_1205_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1205_cast_fp16 = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = query_3_cast_fp16)[name = string("op_1205_cast_fp16")]; tensor var_1209_begin_0 = const()[name = string("op_1209_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1209_end_0 = const()[name = string("op_1209_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_1209_end_mask_0 = const()[name = string("op_1209_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1209_cast_fp16 = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = query_3_cast_fp16)[name = string("op_1209_cast_fp16")]; tensor var_1213_begin_0 = const()[name = string("op_1213_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_1213_end_0 = const()[name = string("op_1213_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_1213_end_mask_0 = const()[name = string("op_1213_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1213_cast_fp16 = slice_by_index(begin = var_1213_begin_0, end = var_1213_end_0, end_mask = var_1213_end_mask_0, x = query_3_cast_fp16)[name = string("op_1213_cast_fp16")]; tensor var_1217_begin_0 = const()[name = string("op_1217_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_1217_end_0 = const()[name = string("op_1217_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_1217_end_mask_0 = const()[name = string("op_1217_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1217_cast_fp16 = slice_by_index(begin = var_1217_begin_0, end = var_1217_end_0, end_mask = var_1217_end_mask_0, x = query_3_cast_fp16)[name = string("op_1217_cast_fp16")]; tensor var_1221_begin_0 = const()[name = string("op_1221_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_1221_end_0 = const()[name = string("op_1221_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_1221_end_mask_0 = const()[name = string("op_1221_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1221_cast_fp16 = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = query_3_cast_fp16)[name = string("op_1221_cast_fp16")]; tensor var_1225_begin_0 = const()[name = string("op_1225_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_1225_end_0 = const()[name = string("op_1225_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_1225_end_mask_0 = const()[name = string("op_1225_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1225_cast_fp16 = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = query_3_cast_fp16)[name = string("op_1225_cast_fp16")]; tensor var_1229_begin_0 = const()[name = string("op_1229_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_1229_end_0 = const()[name = string("op_1229_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_1229_end_mask_0 = const()[name = string("op_1229_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1229_cast_fp16 = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, x = query_3_cast_fp16)[name = string("op_1229_cast_fp16")]; tensor var_1233_begin_0 = const()[name = string("op_1233_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_1233_end_0 = const()[name = string("op_1233_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_1233_end_mask_0 = const()[name = string("op_1233_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1233_cast_fp16 = slice_by_index(begin = var_1233_begin_0, end = var_1233_end_0, end_mask = var_1233_end_mask_0, x = query_3_cast_fp16)[name = string("op_1233_cast_fp16")]; tensor var_1242_begin_0 = const()[name = string("op_1242_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1242_end_0 = const()[name = string("op_1242_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1242_end_mask_0 = const()[name = string("op_1242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1242_cast_fp16 = slice_by_index(begin = var_1242_begin_0, end = var_1242_end_0, end_mask = var_1242_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1242_cast_fp16")]; tensor var_1249_begin_0 = const()[name = string("op_1249_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1249_end_0 = const()[name = string("op_1249_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1249_end_mask_0 = const()[name = string("op_1249_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1249_cast_fp16 = slice_by_index(begin = var_1249_begin_0, end = var_1249_end_0, end_mask = var_1249_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1249_cast_fp16")]; tensor var_1256_begin_0 = const()[name = string("op_1256_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1256_end_0 = const()[name = string("op_1256_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1256_end_mask_0 = const()[name = string("op_1256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1256_cast_fp16 = slice_by_index(begin = var_1256_begin_0, end = var_1256_end_0, end_mask = var_1256_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1256_cast_fp16")]; tensor var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1263_cast_fp16")]; tensor var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1270_cast_fp16")]; tensor var_1277_begin_0 = const()[name = string("op_1277_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1277_end_0 = const()[name = string("op_1277_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1277_end_mask_0 = const()[name = string("op_1277_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1277_cast_fp16 = slice_by_index(begin = var_1277_begin_0, end = var_1277_end_0, end_mask = var_1277_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1277_cast_fp16")]; tensor var_1284_begin_0 = const()[name = string("op_1284_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1284_end_0 = const()[name = string("op_1284_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1284_end_mask_0 = const()[name = string("op_1284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1284_cast_fp16 = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1284_cast_fp16")]; tensor var_1291_begin_0 = const()[name = string("op_1291_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1291_end_0 = const()[name = string("op_1291_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1291_end_mask_0 = const()[name = string("op_1291_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1291_cast_fp16 = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1291_cast_fp16")]; tensor var_1298_begin_0 = const()[name = string("op_1298_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1298_end_0 = const()[name = string("op_1298_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1298_end_mask_0 = const()[name = string("op_1298_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1298_cast_fp16 = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1298_cast_fp16")]; tensor var_1305_begin_0 = const()[name = string("op_1305_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1305_end_0 = const()[name = string("op_1305_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1305_end_mask_0 = const()[name = string("op_1305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1305_cast_fp16 = slice_by_index(begin = var_1305_begin_0, end = var_1305_end_0, end_mask = var_1305_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1305_cast_fp16")]; tensor var_1312_begin_0 = const()[name = string("op_1312_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1312_end_0 = const()[name = string("op_1312_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1312_end_mask_0 = const()[name = string("op_1312_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1312_cast_fp16 = slice_by_index(begin = var_1312_begin_0, end = var_1312_end_0, end_mask = var_1312_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1312_cast_fp16")]; tensor var_1319_begin_0 = const()[name = string("op_1319_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1319_end_0 = const()[name = string("op_1319_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1319_end_mask_0 = const()[name = string("op_1319_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1319_cast_fp16 = slice_by_index(begin = var_1319_begin_0, end = var_1319_end_0, end_mask = var_1319_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1319_cast_fp16")]; tensor var_1326_begin_0 = const()[name = string("op_1326_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1326_end_0 = const()[name = string("op_1326_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1326_end_mask_0 = const()[name = string("op_1326_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1326_cast_fp16 = slice_by_index(begin = var_1326_begin_0, end = var_1326_end_0, end_mask = var_1326_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1326_cast_fp16")]; tensor var_1333_begin_0 = const()[name = string("op_1333_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1333_end_0 = const()[name = string("op_1333_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1333_end_mask_0 = const()[name = string("op_1333_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1333_cast_fp16 = slice_by_index(begin = var_1333_begin_0, end = var_1333_end_0, end_mask = var_1333_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1333_cast_fp16")]; tensor var_1340_begin_0 = const()[name = string("op_1340_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1340_end_0 = const()[name = string("op_1340_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1340_end_mask_0 = const()[name = string("op_1340_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1340_cast_fp16 = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1340_cast_fp16")]; tensor var_1347_begin_0 = const()[name = string("op_1347_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1347_end_0 = const()[name = string("op_1347_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1347_end_mask_0 = const()[name = string("op_1347_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1347_cast_fp16 = slice_by_index(begin = var_1347_begin_0, end = var_1347_end_0, end_mask = var_1347_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1347_cast_fp16")]; tensor var_1354_begin_0 = const()[name = string("op_1354_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1354_end_0 = const()[name = string("op_1354_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1354_end_mask_0 = const()[name = string("op_1354_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1354_cast_fp16 = slice_by_index(begin = var_1354_begin_0, end = var_1354_end_0, end_mask = var_1354_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1354_cast_fp16")]; tensor var_1361_begin_0 = const()[name = string("op_1361_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1361_end_0 = const()[name = string("op_1361_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1361_end_mask_0 = const()[name = string("op_1361_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1361_cast_fp16 = slice_by_index(begin = var_1361_begin_0, end = var_1361_end_0, end_mask = var_1361_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1361_cast_fp16")]; tensor var_1368_begin_0 = const()[name = string("op_1368_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1368_end_0 = const()[name = string("op_1368_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1368_end_mask_0 = const()[name = string("op_1368_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1368_cast_fp16 = slice_by_index(begin = var_1368_begin_0, end = var_1368_end_0, end_mask = var_1368_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1368_cast_fp16")]; tensor var_1375_begin_0 = const()[name = string("op_1375_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1375_end_0 = const()[name = string("op_1375_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1375_end_mask_0 = const()[name = string("op_1375_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1375_cast_fp16 = slice_by_index(begin = var_1375_begin_0, end = var_1375_end_0, end_mask = var_1375_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1375_cast_fp16")]; tensor var_1382_begin_0 = const()[name = string("op_1382_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1382_end_0 = const()[name = string("op_1382_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1382_end_mask_0 = const()[name = string("op_1382_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1382_cast_fp16 = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1382_cast_fp16")]; tensor var_1389_begin_0 = const()[name = string("op_1389_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1389_end_0 = const()[name = string("op_1389_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1389_end_mask_0 = const()[name = string("op_1389_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1389_cast_fp16 = slice_by_index(begin = var_1389_begin_0, end = var_1389_end_0, end_mask = var_1389_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1389_cast_fp16")]; tensor var_1396_begin_0 = const()[name = string("op_1396_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1396_end_0 = const()[name = string("op_1396_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1396_end_mask_0 = const()[name = string("op_1396_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1396_cast_fp16 = slice_by_index(begin = var_1396_begin_0, end = var_1396_end_0, end_mask = var_1396_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1396_cast_fp16")]; tensor var_1403_begin_0 = const()[name = string("op_1403_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1403_end_0 = const()[name = string("op_1403_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1403_end_mask_0 = const()[name = string("op_1403_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1403_cast_fp16 = slice_by_index(begin = var_1403_begin_0, end = var_1403_end_0, end_mask = var_1403_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1403_cast_fp16")]; tensor var_1410_begin_0 = const()[name = string("op_1410_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1410_end_0 = const()[name = string("op_1410_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1410_end_mask_0 = const()[name = string("op_1410_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1410_cast_fp16 = slice_by_index(begin = var_1410_begin_0, end = var_1410_end_0, end_mask = var_1410_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1410_cast_fp16")]; tensor var_1417_begin_0 = const()[name = string("op_1417_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1417_end_0 = const()[name = string("op_1417_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1417_end_mask_0 = const()[name = string("op_1417_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1417_cast_fp16 = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1417_cast_fp16")]; tensor var_1424_begin_0 = const()[name = string("op_1424_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1424_end_0 = const()[name = string("op_1424_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1424_end_mask_0 = const()[name = string("op_1424_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1424_cast_fp16 = slice_by_index(begin = var_1424_begin_0, end = var_1424_end_0, end_mask = var_1424_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1424_cast_fp16")]; tensor var_1431_begin_0 = const()[name = string("op_1431_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1431_end_0 = const()[name = string("op_1431_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1431_end_mask_0 = const()[name = string("op_1431_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1431_cast_fp16 = slice_by_index(begin = var_1431_begin_0, end = var_1431_end_0, end_mask = var_1431_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1431_cast_fp16")]; tensor var_1438_begin_0 = const()[name = string("op_1438_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1438_end_0 = const()[name = string("op_1438_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1438_end_mask_0 = const()[name = string("op_1438_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1438_cast_fp16 = slice_by_index(begin = var_1438_begin_0, end = var_1438_end_0, end_mask = var_1438_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1438_cast_fp16")]; tensor var_1445_begin_0 = const()[name = string("op_1445_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1445_end_0 = const()[name = string("op_1445_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1445_end_mask_0 = const()[name = string("op_1445_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1445_cast_fp16 = slice_by_index(begin = var_1445_begin_0, end = var_1445_end_0, end_mask = var_1445_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1445_cast_fp16")]; tensor var_1452_begin_0 = const()[name = string("op_1452_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1452_end_0 = const()[name = string("op_1452_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1452_end_mask_0 = const()[name = string("op_1452_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1452_cast_fp16 = slice_by_index(begin = var_1452_begin_0, end = var_1452_end_0, end_mask = var_1452_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1452_cast_fp16")]; tensor var_1459_begin_0 = const()[name = string("op_1459_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1459_end_0 = const()[name = string("op_1459_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1459_end_mask_0 = const()[name = string("op_1459_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1459_cast_fp16 = slice_by_index(begin = var_1459_begin_0, end = var_1459_end_0, end_mask = var_1459_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1459_cast_fp16")]; tensor var_1466_begin_0 = const()[name = string("op_1466_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1466_end_0 = const()[name = string("op_1466_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1466_end_mask_0 = const()[name = string("op_1466_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1466_cast_fp16 = slice_by_index(begin = var_1466_begin_0, end = var_1466_end_0, end_mask = var_1466_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1466_cast_fp16")]; tensor var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1473_cast_fp16")]; tensor var_1480_begin_0 = const()[name = string("op_1480_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1480_end_0 = const()[name = string("op_1480_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1480_end_mask_0 = const()[name = string("op_1480_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1480_cast_fp16 = slice_by_index(begin = var_1480_begin_0, end = var_1480_end_0, end_mask = var_1480_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1480_cast_fp16")]; tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1487_cast_fp16")]; tensor var_1494_begin_0 = const()[name = string("op_1494_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1494_end_0 = const()[name = string("op_1494_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1494_end_mask_0 = const()[name = string("op_1494_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1494_cast_fp16 = slice_by_index(begin = var_1494_begin_0, end = var_1494_end_0, end_mask = var_1494_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1494_cast_fp16")]; tensor var_1501_begin_0 = const()[name = string("op_1501_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1501_end_0 = const()[name = string("op_1501_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1501_end_mask_0 = const()[name = string("op_1501_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1501_cast_fp16 = slice_by_index(begin = var_1501_begin_0, end = var_1501_end_0, end_mask = var_1501_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1501_cast_fp16")]; tensor var_1508_begin_0 = const()[name = string("op_1508_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1508_end_0 = const()[name = string("op_1508_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1508_end_mask_0 = const()[name = string("op_1508_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1508_cast_fp16 = slice_by_index(begin = var_1508_begin_0, end = var_1508_end_0, end_mask = var_1508_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1508_cast_fp16")]; tensor var_1515_begin_0 = const()[name = string("op_1515_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1515_end_0 = const()[name = string("op_1515_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1515_end_mask_0 = const()[name = string("op_1515_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1515_cast_fp16 = slice_by_index(begin = var_1515_begin_0, end = var_1515_end_0, end_mask = var_1515_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1515_cast_fp16")]; tensor var_1522_begin_0 = const()[name = string("op_1522_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1522_end_0 = const()[name = string("op_1522_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1522_end_mask_0 = const()[name = string("op_1522_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1522_cast_fp16 = slice_by_index(begin = var_1522_begin_0, end = var_1522_end_0, end_mask = var_1522_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1522_cast_fp16")]; tensor var_1529_begin_0 = const()[name = string("op_1529_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1529_end_0 = const()[name = string("op_1529_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1529_end_mask_0 = const()[name = string("op_1529_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1529_cast_fp16 = slice_by_index(begin = var_1529_begin_0, end = var_1529_end_0, end_mask = var_1529_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1529_cast_fp16")]; tensor var_1536_begin_0 = const()[name = string("op_1536_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1536_end_0 = const()[name = string("op_1536_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1536_end_mask_0 = const()[name = string("op_1536_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1536_cast_fp16 = slice_by_index(begin = var_1536_begin_0, end = var_1536_end_0, end_mask = var_1536_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1536_cast_fp16")]; tensor var_1543_begin_0 = const()[name = string("op_1543_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1543_end_0 = const()[name = string("op_1543_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1543_end_mask_0 = const()[name = string("op_1543_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1543_cast_fp16 = slice_by_index(begin = var_1543_begin_0, end = var_1543_end_0, end_mask = var_1543_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1543_cast_fp16")]; tensor var_1550_begin_0 = const()[name = string("op_1550_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1550_end_0 = const()[name = string("op_1550_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1550_end_mask_0 = const()[name = string("op_1550_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1550_cast_fp16 = slice_by_index(begin = var_1550_begin_0, end = var_1550_end_0, end_mask = var_1550_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1550_cast_fp16")]; tensor var_1557_begin_0 = const()[name = string("op_1557_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1557_end_0 = const()[name = string("op_1557_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1557_end_mask_0 = const()[name = string("op_1557_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1557_cast_fp16 = slice_by_index(begin = var_1557_begin_0, end = var_1557_end_0, end_mask = var_1557_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1557_cast_fp16")]; tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1564_cast_fp16")]; tensor var_1571_begin_0 = const()[name = string("op_1571_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1571_end_0 = const()[name = string("op_1571_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1571_end_mask_0 = const()[name = string("op_1571_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1571_cast_fp16 = slice_by_index(begin = var_1571_begin_0, end = var_1571_end_0, end_mask = var_1571_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1571_cast_fp16")]; tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_10")]; tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = k_3_cast_fp16)[name = string("op_1576_cast_fp16")]; tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = k_3_cast_fp16)[name = string("op_1580_cast_fp16")]; tensor var_1584_begin_0 = const()[name = string("op_1584_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_1584_end_0 = const()[name = string("op_1584_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_1584_end_mask_0 = const()[name = string("op_1584_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = k_3_cast_fp16)[name = string("op_1584_cast_fp16")]; tensor var_1588_begin_0 = const()[name = string("op_1588_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_1588_end_0 = const()[name = string("op_1588_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_1588_end_mask_0 = const()[name = string("op_1588_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1588_cast_fp16 = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = k_3_cast_fp16)[name = string("op_1588_cast_fp16")]; tensor var_1592_begin_0 = const()[name = string("op_1592_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_1592_end_0 = const()[name = string("op_1592_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_1592_end_mask_0 = const()[name = string("op_1592_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1592_cast_fp16 = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = k_3_cast_fp16)[name = string("op_1592_cast_fp16")]; tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = k_3_cast_fp16)[name = string("op_1596_cast_fp16")]; tensor var_1600_begin_0 = const()[name = string("op_1600_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_1600_end_0 = const()[name = string("op_1600_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_1600_end_mask_0 = const()[name = string("op_1600_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, x = k_3_cast_fp16)[name = string("op_1600_cast_fp16")]; tensor var_1604_begin_0 = const()[name = string("op_1604_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_1604_end_0 = const()[name = string("op_1604_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_1604_end_mask_0 = const()[name = string("op_1604_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1604_cast_fp16 = slice_by_index(begin = var_1604_begin_0, end = var_1604_end_0, end_mask = var_1604_end_mask_0, x = k_3_cast_fp16)[name = string("op_1604_cast_fp16")]; tensor var_1608_begin_0 = const()[name = string("op_1608_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1608_end_0 = const()[name = string("op_1608_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_1608_end_mask_0 = const()[name = string("op_1608_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1608_cast_fp16 = slice_by_index(begin = var_1608_begin_0, end = var_1608_end_0, end_mask = var_1608_end_mask_0, x = k_3_cast_fp16)[name = string("op_1608_cast_fp16")]; tensor var_1612_begin_0 = const()[name = string("op_1612_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_1612_end_0 = const()[name = string("op_1612_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_1612_end_mask_0 = const()[name = string("op_1612_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1612_cast_fp16 = slice_by_index(begin = var_1612_begin_0, end = var_1612_end_0, end_mask = var_1612_end_mask_0, x = k_3_cast_fp16)[name = string("op_1612_cast_fp16")]; tensor var_1616_begin_0 = const()[name = string("op_1616_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_1616_end_0 = const()[name = string("op_1616_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_1616_end_mask_0 = const()[name = string("op_1616_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1616_cast_fp16 = slice_by_index(begin = var_1616_begin_0, end = var_1616_end_0, end_mask = var_1616_end_mask_0, x = k_3_cast_fp16)[name = string("op_1616_cast_fp16")]; tensor var_1620_begin_0 = const()[name = string("op_1620_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_1620_end_0 = const()[name = string("op_1620_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_1620_end_mask_0 = const()[name = string("op_1620_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1620_cast_fp16 = slice_by_index(begin = var_1620_begin_0, end = var_1620_end_0, end_mask = var_1620_end_mask_0, x = k_3_cast_fp16)[name = string("op_1620_cast_fp16")]; tensor var_1622_begin_0 = const()[name = string("op_1622_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1622_end_0 = const()[name = string("op_1622_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1622_end_mask_0 = const()[name = string("op_1622_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1622_cast_fp16 = slice_by_index(begin = var_1622_begin_0, end = var_1622_end_0, end_mask = var_1622_end_mask_0, x = value_3_cast_fp16)[name = string("op_1622_cast_fp16")]; tensor var_1626_begin_0 = const()[name = string("op_1626_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_1626_end_0 = const()[name = string("op_1626_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_1626_end_mask_0 = const()[name = string("op_1626_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1626_cast_fp16 = slice_by_index(begin = var_1626_begin_0, end = var_1626_end_0, end_mask = var_1626_end_mask_0, x = value_3_cast_fp16)[name = string("op_1626_cast_fp16")]; tensor var_1630_begin_0 = const()[name = string("op_1630_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_1630_end_0 = const()[name = string("op_1630_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_1630_end_mask_0 = const()[name = string("op_1630_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1630_cast_fp16 = slice_by_index(begin = var_1630_begin_0, end = var_1630_end_0, end_mask = var_1630_end_mask_0, x = value_3_cast_fp16)[name = string("op_1630_cast_fp16")]; tensor var_1634_begin_0 = const()[name = string("op_1634_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_1634_end_0 = const()[name = string("op_1634_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_1634_end_mask_0 = const()[name = string("op_1634_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1634_cast_fp16 = slice_by_index(begin = var_1634_begin_0, end = var_1634_end_0, end_mask = var_1634_end_mask_0, x = value_3_cast_fp16)[name = string("op_1634_cast_fp16")]; tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = value_3_cast_fp16)[name = string("op_1638_cast_fp16")]; tensor var_1642_begin_0 = const()[name = string("op_1642_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1642_end_0 = const()[name = string("op_1642_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_1642_end_mask_0 = const()[name = string("op_1642_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1642_cast_fp16 = slice_by_index(begin = var_1642_begin_0, end = var_1642_end_0, end_mask = var_1642_end_mask_0, x = value_3_cast_fp16)[name = string("op_1642_cast_fp16")]; tensor var_1646_begin_0 = const()[name = string("op_1646_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_1646_end_0 = const()[name = string("op_1646_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_1646_end_mask_0 = const()[name = string("op_1646_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1646_cast_fp16 = slice_by_index(begin = var_1646_begin_0, end = var_1646_end_0, end_mask = var_1646_end_mask_0, x = value_3_cast_fp16)[name = string("op_1646_cast_fp16")]; tensor var_1650_begin_0 = const()[name = string("op_1650_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_1650_end_0 = const()[name = string("op_1650_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_1650_end_mask_0 = const()[name = string("op_1650_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1650_cast_fp16 = slice_by_index(begin = var_1650_begin_0, end = var_1650_end_0, end_mask = var_1650_end_mask_0, x = value_3_cast_fp16)[name = string("op_1650_cast_fp16")]; tensor var_1654_begin_0 = const()[name = string("op_1654_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_1654_end_0 = const()[name = string("op_1654_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_1654_end_mask_0 = const()[name = string("op_1654_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1654_cast_fp16 = slice_by_index(begin = var_1654_begin_0, end = var_1654_end_0, end_mask = var_1654_end_mask_0, x = value_3_cast_fp16)[name = string("op_1654_cast_fp16")]; tensor var_1658_begin_0 = const()[name = string("op_1658_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_1658_end_0 = const()[name = string("op_1658_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_1658_end_mask_0 = const()[name = string("op_1658_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1658_cast_fp16 = slice_by_index(begin = var_1658_begin_0, end = var_1658_end_0, end_mask = var_1658_end_mask_0, x = value_3_cast_fp16)[name = string("op_1658_cast_fp16")]; tensor var_1662_begin_0 = const()[name = string("op_1662_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_1662_end_0 = const()[name = string("op_1662_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_1662_end_mask_0 = const()[name = string("op_1662_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1662_cast_fp16 = slice_by_index(begin = var_1662_begin_0, end = var_1662_end_0, end_mask = var_1662_end_mask_0, x = value_3_cast_fp16)[name = string("op_1662_cast_fp16")]; tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = value_3_cast_fp16)[name = string("op_1666_cast_fp16")]; string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1576_cast_fp16, var_1242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1576_cast_fp16, var_1249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1576_cast_fp16, var_1256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1576_cast_fp16, var_1263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1580_cast_fp16, var_1270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1580_cast_fp16, var_1277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1580_cast_fp16, var_1284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1580_cast_fp16, var_1291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1584_cast_fp16, var_1298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1584_cast_fp16, var_1305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1584_cast_fp16, var_1312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1584_cast_fp16, var_1319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1588_cast_fp16, var_1326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1588_cast_fp16, var_1333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1588_cast_fp16, var_1340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1588_cast_fp16, var_1347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1592_cast_fp16, var_1354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1592_cast_fp16, var_1361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1592_cast_fp16, var_1368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1592_cast_fp16, var_1375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1596_cast_fp16, var_1382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1596_cast_fp16, var_1389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1596_cast_fp16, var_1396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1596_cast_fp16, var_1403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1600_cast_fp16, var_1410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1600_cast_fp16, var_1417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1600_cast_fp16, var_1424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1600_cast_fp16, var_1431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1604_cast_fp16, var_1438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1604_cast_fp16, var_1445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1604_cast_fp16, var_1452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1604_cast_fp16, var_1459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1608_cast_fp16, var_1466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1608_cast_fp16, var_1473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1608_cast_fp16, var_1480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1608_cast_fp16, var_1487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1612_cast_fp16, var_1494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1612_cast_fp16, var_1501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1612_cast_fp16, var_1508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1612_cast_fp16, var_1515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1616_cast_fp16, var_1522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1616_cast_fp16, var_1529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1616_cast_fp16, var_1536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1616_cast_fp16, var_1543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1620_cast_fp16, var_1550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1620_cast_fp16, var_1557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1620_cast_fp16, var_1564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1620_cast_fp16, var_1571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; fp16 var_1765_to_fp16 = const()[name = string("op_1765_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1765_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; fp16 var_1767_to_fp16 = const()[name = string("op_1767_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1767_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; fp16 var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1769_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; fp16 var_1771_to_fp16 = const()[name = string("op_1771_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1771_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; fp16 var_1773_to_fp16 = const()[name = string("op_1773_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1773_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; fp16 var_1775_to_fp16 = const()[name = string("op_1775_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1775_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; fp16 var_1777_to_fp16 = const()[name = string("op_1777_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1777_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; fp16 var_1779_to_fp16 = const()[name = string("op_1779_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1779_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; fp16 var_1781_to_fp16 = const()[name = string("op_1781_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1781_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; fp16 var_1783_to_fp16 = const()[name = string("op_1783_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1783_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; fp16 var_1785_to_fp16 = const()[name = string("op_1785_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1785_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; fp16 var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1787_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; fp16 var_1789_to_fp16 = const()[name = string("op_1789_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1789_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; fp16 var_1791_to_fp16 = const()[name = string("op_1791_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1791_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; fp16 var_1793_to_fp16 = const()[name = string("op_1793_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1793_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; fp16 var_1795_to_fp16 = const()[name = string("op_1795_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1795_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; fp16 var_1797_to_fp16 = const()[name = string("op_1797_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1797_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; fp16 var_1799_to_fp16 = const()[name = string("op_1799_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1799_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; fp16 var_1801_to_fp16 = const()[name = string("op_1801_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1801_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; fp16 var_1803_to_fp16 = const()[name = string("op_1803_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1803_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1805_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; fp16 var_1807_to_fp16 = const()[name = string("op_1807_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1807_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; fp16 var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1809_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; fp16 var_1811_to_fp16 = const()[name = string("op_1811_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1811_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; fp16 var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1813_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; fp16 var_1815_to_fp16 = const()[name = string("op_1815_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1815_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; fp16 var_1817_to_fp16 = const()[name = string("op_1817_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1817_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; fp16 var_1819_to_fp16 = const()[name = string("op_1819_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1819_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; fp16 var_1821_to_fp16 = const()[name = string("op_1821_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1821_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; fp16 var_1823_to_fp16 = const()[name = string("op_1823_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1823_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; fp16 var_1825_to_fp16 = const()[name = string("op_1825_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1825_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; fp16 var_1827_to_fp16 = const()[name = string("op_1827_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1827_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; fp16 var_1829_to_fp16 = const()[name = string("op_1829_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1829_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; fp16 var_1831_to_fp16 = const()[name = string("op_1831_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1831_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1833_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; fp16 var_1835_to_fp16 = const()[name = string("op_1835_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1835_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; fp16 var_1837_to_fp16 = const()[name = string("op_1837_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1837_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; fp16 var_1839_to_fp16 = const()[name = string("op_1839_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1839_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; fp16 var_1841_to_fp16 = const()[name = string("op_1841_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1841_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; fp16 var_1843_to_fp16 = const()[name = string("op_1843_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1843_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; fp16 var_1845_to_fp16 = const()[name = string("op_1845_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1845_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; fp16 var_1847_to_fp16 = const()[name = string("op_1847_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1847_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; fp16 var_1849_to_fp16 = const()[name = string("op_1849_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1849_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; fp16 var_1851_to_fp16 = const()[name = string("op_1851_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1851_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; fp16 var_1853_to_fp16 = const()[name = string("op_1853_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1853_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; fp16 var_1855_to_fp16 = const()[name = string("op_1855_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1855_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; fp16 var_1857_to_fp16 = const()[name = string("op_1857_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1857_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; fp16 var_1859_to_fp16 = const()[name = string("op_1859_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1859_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; tensor var_1861_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_97_cast_fp16)[name = string("op_1861_cast_fp16")]; tensor var_1862_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_99_cast_fp16)[name = string("op_1862_cast_fp16")]; tensor var_1863_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_101_cast_fp16)[name = string("op_1863_cast_fp16")]; tensor var_1864_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_103_cast_fp16)[name = string("op_1864_cast_fp16")]; tensor var_1865_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_105_cast_fp16)[name = string("op_1865_cast_fp16")]; tensor var_1866_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_107_cast_fp16)[name = string("op_1866_cast_fp16")]; tensor var_1867_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_109_cast_fp16)[name = string("op_1867_cast_fp16")]; tensor var_1868_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_111_cast_fp16)[name = string("op_1868_cast_fp16")]; tensor var_1869_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_113_cast_fp16)[name = string("op_1869_cast_fp16")]; tensor var_1870_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_115_cast_fp16)[name = string("op_1870_cast_fp16")]; tensor var_1871_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_117_cast_fp16)[name = string("op_1871_cast_fp16")]; tensor var_1872_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_119_cast_fp16)[name = string("op_1872_cast_fp16")]; tensor var_1873_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_121_cast_fp16)[name = string("op_1873_cast_fp16")]; tensor var_1874_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_123_cast_fp16)[name = string("op_1874_cast_fp16")]; tensor var_1875_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_125_cast_fp16)[name = string("op_1875_cast_fp16")]; tensor var_1876_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_127_cast_fp16)[name = string("op_1876_cast_fp16")]; tensor var_1877_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_129_cast_fp16)[name = string("op_1877_cast_fp16")]; tensor var_1878_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_131_cast_fp16)[name = string("op_1878_cast_fp16")]; tensor var_1879_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_133_cast_fp16)[name = string("op_1879_cast_fp16")]; tensor var_1880_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_135_cast_fp16)[name = string("op_1880_cast_fp16")]; tensor var_1881_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_137_cast_fp16)[name = string("op_1881_cast_fp16")]; tensor var_1882_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_139_cast_fp16)[name = string("op_1882_cast_fp16")]; tensor var_1883_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_141_cast_fp16)[name = string("op_1883_cast_fp16")]; tensor var_1884_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_143_cast_fp16)[name = string("op_1884_cast_fp16")]; tensor var_1885_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_145_cast_fp16)[name = string("op_1885_cast_fp16")]; tensor var_1886_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_147_cast_fp16)[name = string("op_1886_cast_fp16")]; tensor var_1887_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_149_cast_fp16)[name = string("op_1887_cast_fp16")]; tensor var_1888_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_151_cast_fp16)[name = string("op_1888_cast_fp16")]; tensor var_1889_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_153_cast_fp16)[name = string("op_1889_cast_fp16")]; tensor var_1890_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_155_cast_fp16)[name = string("op_1890_cast_fp16")]; tensor var_1891_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_157_cast_fp16)[name = string("op_1891_cast_fp16")]; tensor var_1892_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_159_cast_fp16)[name = string("op_1892_cast_fp16")]; tensor var_1893_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_161_cast_fp16)[name = string("op_1893_cast_fp16")]; tensor var_1894_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_163_cast_fp16)[name = string("op_1894_cast_fp16")]; tensor var_1895_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_165_cast_fp16)[name = string("op_1895_cast_fp16")]; tensor var_1896_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_167_cast_fp16)[name = string("op_1896_cast_fp16")]; tensor var_1897_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_169_cast_fp16)[name = string("op_1897_cast_fp16")]; tensor var_1898_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_171_cast_fp16)[name = string("op_1898_cast_fp16")]; tensor var_1899_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_173_cast_fp16)[name = string("op_1899_cast_fp16")]; tensor var_1900_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_175_cast_fp16)[name = string("op_1900_cast_fp16")]; tensor var_1901_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_177_cast_fp16)[name = string("op_1901_cast_fp16")]; tensor var_1902_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_179_cast_fp16)[name = string("op_1902_cast_fp16")]; tensor var_1903_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_181_cast_fp16)[name = string("op_1903_cast_fp16")]; tensor var_1904_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_183_cast_fp16)[name = string("op_1904_cast_fp16")]; tensor var_1905_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_185_cast_fp16)[name = string("op_1905_cast_fp16")]; tensor var_1906_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_187_cast_fp16)[name = string("op_1906_cast_fp16")]; tensor var_1907_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_189_cast_fp16)[name = string("op_1907_cast_fp16")]; tensor var_1908_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_191_cast_fp16)[name = string("op_1908_cast_fp16")]; string var_1910_equation_0 = const()[name = string("op_1910_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1910_cast_fp16 = einsum(equation = var_1910_equation_0, values = (var_1622_cast_fp16, var_1861_cast_fp16))[name = string("op_1910_cast_fp16")]; string var_1912_equation_0 = const()[name = string("op_1912_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1912_cast_fp16 = einsum(equation = var_1912_equation_0, values = (var_1622_cast_fp16, var_1862_cast_fp16))[name = string("op_1912_cast_fp16")]; string var_1914_equation_0 = const()[name = string("op_1914_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1914_cast_fp16 = einsum(equation = var_1914_equation_0, values = (var_1622_cast_fp16, var_1863_cast_fp16))[name = string("op_1914_cast_fp16")]; string var_1916_equation_0 = const()[name = string("op_1916_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1916_cast_fp16 = einsum(equation = var_1916_equation_0, values = (var_1622_cast_fp16, var_1864_cast_fp16))[name = string("op_1916_cast_fp16")]; string var_1918_equation_0 = const()[name = string("op_1918_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1918_cast_fp16 = einsum(equation = var_1918_equation_0, values = (var_1626_cast_fp16, var_1865_cast_fp16))[name = string("op_1918_cast_fp16")]; string var_1920_equation_0 = const()[name = string("op_1920_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1920_cast_fp16 = einsum(equation = var_1920_equation_0, values = (var_1626_cast_fp16, var_1866_cast_fp16))[name = string("op_1920_cast_fp16")]; string var_1922_equation_0 = const()[name = string("op_1922_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1922_cast_fp16 = einsum(equation = var_1922_equation_0, values = (var_1626_cast_fp16, var_1867_cast_fp16))[name = string("op_1922_cast_fp16")]; string var_1924_equation_0 = const()[name = string("op_1924_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1924_cast_fp16 = einsum(equation = var_1924_equation_0, values = (var_1626_cast_fp16, var_1868_cast_fp16))[name = string("op_1924_cast_fp16")]; string var_1926_equation_0 = const()[name = string("op_1926_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1926_cast_fp16 = einsum(equation = var_1926_equation_0, values = (var_1630_cast_fp16, var_1869_cast_fp16))[name = string("op_1926_cast_fp16")]; string var_1928_equation_0 = const()[name = string("op_1928_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1928_cast_fp16 = einsum(equation = var_1928_equation_0, values = (var_1630_cast_fp16, var_1870_cast_fp16))[name = string("op_1928_cast_fp16")]; string var_1930_equation_0 = const()[name = string("op_1930_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1930_cast_fp16 = einsum(equation = var_1930_equation_0, values = (var_1630_cast_fp16, var_1871_cast_fp16))[name = string("op_1930_cast_fp16")]; string var_1932_equation_0 = const()[name = string("op_1932_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1932_cast_fp16 = einsum(equation = var_1932_equation_0, values = (var_1630_cast_fp16, var_1872_cast_fp16))[name = string("op_1932_cast_fp16")]; string var_1934_equation_0 = const()[name = string("op_1934_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1934_cast_fp16 = einsum(equation = var_1934_equation_0, values = (var_1634_cast_fp16, var_1873_cast_fp16))[name = string("op_1934_cast_fp16")]; string var_1936_equation_0 = const()[name = string("op_1936_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1936_cast_fp16 = einsum(equation = var_1936_equation_0, values = (var_1634_cast_fp16, var_1874_cast_fp16))[name = string("op_1936_cast_fp16")]; string var_1938_equation_0 = const()[name = string("op_1938_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1938_cast_fp16 = einsum(equation = var_1938_equation_0, values = (var_1634_cast_fp16, var_1875_cast_fp16))[name = string("op_1938_cast_fp16")]; string var_1940_equation_0 = const()[name = string("op_1940_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1940_cast_fp16 = einsum(equation = var_1940_equation_0, values = (var_1634_cast_fp16, var_1876_cast_fp16))[name = string("op_1940_cast_fp16")]; string var_1942_equation_0 = const()[name = string("op_1942_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1942_cast_fp16 = einsum(equation = var_1942_equation_0, values = (var_1638_cast_fp16, var_1877_cast_fp16))[name = string("op_1942_cast_fp16")]; string var_1944_equation_0 = const()[name = string("op_1944_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1944_cast_fp16 = einsum(equation = var_1944_equation_0, values = (var_1638_cast_fp16, var_1878_cast_fp16))[name = string("op_1944_cast_fp16")]; string var_1946_equation_0 = const()[name = string("op_1946_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1946_cast_fp16 = einsum(equation = var_1946_equation_0, values = (var_1638_cast_fp16, var_1879_cast_fp16))[name = string("op_1946_cast_fp16")]; string var_1948_equation_0 = const()[name = string("op_1948_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1948_cast_fp16 = einsum(equation = var_1948_equation_0, values = (var_1638_cast_fp16, var_1880_cast_fp16))[name = string("op_1948_cast_fp16")]; string var_1950_equation_0 = const()[name = string("op_1950_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1950_cast_fp16 = einsum(equation = var_1950_equation_0, values = (var_1642_cast_fp16, var_1881_cast_fp16))[name = string("op_1950_cast_fp16")]; string var_1952_equation_0 = const()[name = string("op_1952_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1952_cast_fp16 = einsum(equation = var_1952_equation_0, values = (var_1642_cast_fp16, var_1882_cast_fp16))[name = string("op_1952_cast_fp16")]; string var_1954_equation_0 = const()[name = string("op_1954_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1954_cast_fp16 = einsum(equation = var_1954_equation_0, values = (var_1642_cast_fp16, var_1883_cast_fp16))[name = string("op_1954_cast_fp16")]; string var_1956_equation_0 = const()[name = string("op_1956_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1956_cast_fp16 = einsum(equation = var_1956_equation_0, values = (var_1642_cast_fp16, var_1884_cast_fp16))[name = string("op_1956_cast_fp16")]; string var_1958_equation_0 = const()[name = string("op_1958_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1958_cast_fp16 = einsum(equation = var_1958_equation_0, values = (var_1646_cast_fp16, var_1885_cast_fp16))[name = string("op_1958_cast_fp16")]; string var_1960_equation_0 = const()[name = string("op_1960_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1960_cast_fp16 = einsum(equation = var_1960_equation_0, values = (var_1646_cast_fp16, var_1886_cast_fp16))[name = string("op_1960_cast_fp16")]; string var_1962_equation_0 = const()[name = string("op_1962_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1962_cast_fp16 = einsum(equation = var_1962_equation_0, values = (var_1646_cast_fp16, var_1887_cast_fp16))[name = string("op_1962_cast_fp16")]; string var_1964_equation_0 = const()[name = string("op_1964_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1964_cast_fp16 = einsum(equation = var_1964_equation_0, values = (var_1646_cast_fp16, var_1888_cast_fp16))[name = string("op_1964_cast_fp16")]; string var_1966_equation_0 = const()[name = string("op_1966_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1966_cast_fp16 = einsum(equation = var_1966_equation_0, values = (var_1650_cast_fp16, var_1889_cast_fp16))[name = string("op_1966_cast_fp16")]; string var_1968_equation_0 = const()[name = string("op_1968_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1968_cast_fp16 = einsum(equation = var_1968_equation_0, values = (var_1650_cast_fp16, var_1890_cast_fp16))[name = string("op_1968_cast_fp16")]; string var_1970_equation_0 = const()[name = string("op_1970_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1970_cast_fp16 = einsum(equation = var_1970_equation_0, values = (var_1650_cast_fp16, var_1891_cast_fp16))[name = string("op_1970_cast_fp16")]; string var_1972_equation_0 = const()[name = string("op_1972_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1972_cast_fp16 = einsum(equation = var_1972_equation_0, values = (var_1650_cast_fp16, var_1892_cast_fp16))[name = string("op_1972_cast_fp16")]; string var_1974_equation_0 = const()[name = string("op_1974_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1974_cast_fp16 = einsum(equation = var_1974_equation_0, values = (var_1654_cast_fp16, var_1893_cast_fp16))[name = string("op_1974_cast_fp16")]; string var_1976_equation_0 = const()[name = string("op_1976_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1976_cast_fp16 = einsum(equation = var_1976_equation_0, values = (var_1654_cast_fp16, var_1894_cast_fp16))[name = string("op_1976_cast_fp16")]; string var_1978_equation_0 = const()[name = string("op_1978_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1978_cast_fp16 = einsum(equation = var_1978_equation_0, values = (var_1654_cast_fp16, var_1895_cast_fp16))[name = string("op_1978_cast_fp16")]; string var_1980_equation_0 = const()[name = string("op_1980_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1980_cast_fp16 = einsum(equation = var_1980_equation_0, values = (var_1654_cast_fp16, var_1896_cast_fp16))[name = string("op_1980_cast_fp16")]; string var_1982_equation_0 = const()[name = string("op_1982_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1982_cast_fp16 = einsum(equation = var_1982_equation_0, values = (var_1658_cast_fp16, var_1897_cast_fp16))[name = string("op_1982_cast_fp16")]; string var_1984_equation_0 = const()[name = string("op_1984_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1984_cast_fp16 = einsum(equation = var_1984_equation_0, values = (var_1658_cast_fp16, var_1898_cast_fp16))[name = string("op_1984_cast_fp16")]; string var_1986_equation_0 = const()[name = string("op_1986_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1986_cast_fp16 = einsum(equation = var_1986_equation_0, values = (var_1658_cast_fp16, var_1899_cast_fp16))[name = string("op_1986_cast_fp16")]; string var_1988_equation_0 = const()[name = string("op_1988_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1988_cast_fp16 = einsum(equation = var_1988_equation_0, values = (var_1658_cast_fp16, var_1900_cast_fp16))[name = string("op_1988_cast_fp16")]; string var_1990_equation_0 = const()[name = string("op_1990_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1990_cast_fp16 = einsum(equation = var_1990_equation_0, values = (var_1662_cast_fp16, var_1901_cast_fp16))[name = string("op_1990_cast_fp16")]; string var_1992_equation_0 = const()[name = string("op_1992_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1992_cast_fp16 = einsum(equation = var_1992_equation_0, values = (var_1662_cast_fp16, var_1902_cast_fp16))[name = string("op_1992_cast_fp16")]; string var_1994_equation_0 = const()[name = string("op_1994_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1994_cast_fp16 = einsum(equation = var_1994_equation_0, values = (var_1662_cast_fp16, var_1903_cast_fp16))[name = string("op_1994_cast_fp16")]; string var_1996_equation_0 = const()[name = string("op_1996_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1996_cast_fp16 = einsum(equation = var_1996_equation_0, values = (var_1662_cast_fp16, var_1904_cast_fp16))[name = string("op_1996_cast_fp16")]; string var_1998_equation_0 = const()[name = string("op_1998_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_1998_cast_fp16 = einsum(equation = var_1998_equation_0, values = (var_1666_cast_fp16, var_1905_cast_fp16))[name = string("op_1998_cast_fp16")]; string var_2000_equation_0 = const()[name = string("op_2000_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2000_cast_fp16 = einsum(equation = var_2000_equation_0, values = (var_1666_cast_fp16, var_1906_cast_fp16))[name = string("op_2000_cast_fp16")]; string var_2002_equation_0 = const()[name = string("op_2002_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2002_cast_fp16 = einsum(equation = var_2002_equation_0, values = (var_1666_cast_fp16, var_1907_cast_fp16))[name = string("op_2002_cast_fp16")]; string var_2004_equation_0 = const()[name = string("op_2004_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2004_cast_fp16 = einsum(equation = var_2004_equation_0, values = (var_1666_cast_fp16, var_1908_cast_fp16))[name = string("op_2004_cast_fp16")]; bool var_2006_interleave_0 = const()[name = string("op_2006_interleave_0"), val = bool(false)]; tensor var_2006_cast_fp16 = concat(axis = var_1117, interleave = var_2006_interleave_0, values = (var_1910_cast_fp16, var_1912_cast_fp16, var_1914_cast_fp16, var_1916_cast_fp16))[name = string("op_2006_cast_fp16")]; bool var_2008_interleave_0 = const()[name = string("op_2008_interleave_0"), val = bool(false)]; tensor var_2008_cast_fp16 = concat(axis = var_1117, interleave = var_2008_interleave_0, values = (var_1918_cast_fp16, var_1920_cast_fp16, var_1922_cast_fp16, var_1924_cast_fp16))[name = string("op_2008_cast_fp16")]; bool var_2010_interleave_0 = const()[name = string("op_2010_interleave_0"), val = bool(false)]; tensor var_2010_cast_fp16 = concat(axis = var_1117, interleave = var_2010_interleave_0, values = (var_1926_cast_fp16, var_1928_cast_fp16, var_1930_cast_fp16, var_1932_cast_fp16))[name = string("op_2010_cast_fp16")]; bool var_2012_interleave_0 = const()[name = string("op_2012_interleave_0"), val = bool(false)]; tensor var_2012_cast_fp16 = concat(axis = var_1117, interleave = var_2012_interleave_0, values = (var_1934_cast_fp16, var_1936_cast_fp16, var_1938_cast_fp16, var_1940_cast_fp16))[name = string("op_2012_cast_fp16")]; bool var_2014_interleave_0 = const()[name = string("op_2014_interleave_0"), val = bool(false)]; tensor var_2014_cast_fp16 = concat(axis = var_1117, interleave = var_2014_interleave_0, values = (var_1942_cast_fp16, var_1944_cast_fp16, var_1946_cast_fp16, var_1948_cast_fp16))[name = string("op_2014_cast_fp16")]; bool var_2016_interleave_0 = const()[name = string("op_2016_interleave_0"), val = bool(false)]; tensor var_2016_cast_fp16 = concat(axis = var_1117, interleave = var_2016_interleave_0, values = (var_1950_cast_fp16, var_1952_cast_fp16, var_1954_cast_fp16, var_1956_cast_fp16))[name = string("op_2016_cast_fp16")]; bool var_2018_interleave_0 = const()[name = string("op_2018_interleave_0"), val = bool(false)]; tensor var_2018_cast_fp16 = concat(axis = var_1117, interleave = var_2018_interleave_0, values = (var_1958_cast_fp16, var_1960_cast_fp16, var_1962_cast_fp16, var_1964_cast_fp16))[name = string("op_2018_cast_fp16")]; bool var_2020_interleave_0 = const()[name = string("op_2020_interleave_0"), val = bool(false)]; tensor var_2020_cast_fp16 = concat(axis = var_1117, interleave = var_2020_interleave_0, values = (var_1966_cast_fp16, var_1968_cast_fp16, var_1970_cast_fp16, var_1972_cast_fp16))[name = string("op_2020_cast_fp16")]; bool var_2022_interleave_0 = const()[name = string("op_2022_interleave_0"), val = bool(false)]; tensor var_2022_cast_fp16 = concat(axis = var_1117, interleave = var_2022_interleave_0, values = (var_1974_cast_fp16, var_1976_cast_fp16, var_1978_cast_fp16, var_1980_cast_fp16))[name = string("op_2022_cast_fp16")]; bool var_2024_interleave_0 = const()[name = string("op_2024_interleave_0"), val = bool(false)]; tensor var_2024_cast_fp16 = concat(axis = var_1117, interleave = var_2024_interleave_0, values = (var_1982_cast_fp16, var_1984_cast_fp16, var_1986_cast_fp16, var_1988_cast_fp16))[name = string("op_2024_cast_fp16")]; bool var_2026_interleave_0 = const()[name = string("op_2026_interleave_0"), val = bool(false)]; tensor var_2026_cast_fp16 = concat(axis = var_1117, interleave = var_2026_interleave_0, values = (var_1990_cast_fp16, var_1992_cast_fp16, var_1994_cast_fp16, var_1996_cast_fp16))[name = string("op_2026_cast_fp16")]; bool var_2028_interleave_0 = const()[name = string("op_2028_interleave_0"), val = bool(false)]; tensor var_2028_cast_fp16 = concat(axis = var_1117, interleave = var_2028_interleave_0, values = (var_1998_cast_fp16, var_2000_cast_fp16, var_2002_cast_fp16, var_2004_cast_fp16))[name = string("op_2028_cast_fp16")]; bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; tensor input_9_cast_fp16 = concat(axis = var_1134, interleave = input_9_interleave_0, values = (var_2006_cast_fp16, var_2008_cast_fp16, var_2010_cast_fp16, var_2012_cast_fp16, var_2014_cast_fp16, var_2016_cast_fp16, var_2018_cast_fp16, var_2020_cast_fp16, var_2022_cast_fp16, var_2024_cast_fp16, var_2026_cast_fp16, var_2028_cast_fp16))[name = string("input_9_cast_fp16")]; string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23938944)))]; tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25118656)))]; tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; fp16 var_2047_to_fp16 = const()[name = string("op_2047_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_2047_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25120256)))]; tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25121856)))]; fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25123456)))]; tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29842112)))]; tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29848320)))]; tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34566976)))]; tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; int32 var_2076 = const()[name = string("op_2076"), val = int32(3)]; int32 var_2093 = const()[name = string("op_2093"), val = int32(1)]; tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; fp16 var_2110_to_fp16 = const()[name = string("op_2110_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_2110_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34568576)))]; tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34570176)))]; fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34571776)))]; tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35751488)))]; tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35753088)))]; tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36932800)))]; tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38112512)))]; tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; tensor var_2148_begin_0 = const()[name = string("op_2148_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2148_end_0 = const()[name = string("op_2148_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2148_end_mask_0 = const()[name = string("op_2148_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = query_5_cast_fp16)[name = string("op_2148_cast_fp16")]; tensor var_2152_begin_0 = const()[name = string("op_2152_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_2152_end_0 = const()[name = string("op_2152_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_2152_end_mask_0 = const()[name = string("op_2152_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = query_5_cast_fp16)[name = string("op_2152_cast_fp16")]; tensor var_2156_begin_0 = const()[name = string("op_2156_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_2156_end_0 = const()[name = string("op_2156_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_2156_end_mask_0 = const()[name = string("op_2156_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = query_5_cast_fp16)[name = string("op_2156_cast_fp16")]; tensor var_2160_begin_0 = const()[name = string("op_2160_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_2160_end_0 = const()[name = string("op_2160_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_2160_end_mask_0 = const()[name = string("op_2160_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = query_5_cast_fp16)[name = string("op_2160_cast_fp16")]; tensor var_2164_begin_0 = const()[name = string("op_2164_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_2164_end_0 = const()[name = string("op_2164_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_2164_end_mask_0 = const()[name = string("op_2164_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = query_5_cast_fp16)[name = string("op_2164_cast_fp16")]; tensor var_2168_begin_0 = const()[name = string("op_2168_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2168_end_0 = const()[name = string("op_2168_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_2168_end_mask_0 = const()[name = string("op_2168_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, x = query_5_cast_fp16)[name = string("op_2168_cast_fp16")]; tensor var_2172_begin_0 = const()[name = string("op_2172_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_2172_end_0 = const()[name = string("op_2172_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_2172_end_mask_0 = const()[name = string("op_2172_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2172_cast_fp16 = slice_by_index(begin = var_2172_begin_0, end = var_2172_end_0, end_mask = var_2172_end_mask_0, x = query_5_cast_fp16)[name = string("op_2172_cast_fp16")]; tensor var_2176_begin_0 = const()[name = string("op_2176_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_2176_end_0 = const()[name = string("op_2176_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_2176_end_mask_0 = const()[name = string("op_2176_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2176_cast_fp16 = slice_by_index(begin = var_2176_begin_0, end = var_2176_end_0, end_mask = var_2176_end_mask_0, x = query_5_cast_fp16)[name = string("op_2176_cast_fp16")]; tensor var_2180_begin_0 = const()[name = string("op_2180_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_2180_end_0 = const()[name = string("op_2180_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_2180_end_mask_0 = const()[name = string("op_2180_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2180_cast_fp16 = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, x = query_5_cast_fp16)[name = string("op_2180_cast_fp16")]; tensor var_2184_begin_0 = const()[name = string("op_2184_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_2184_end_0 = const()[name = string("op_2184_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_2184_end_mask_0 = const()[name = string("op_2184_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2184_cast_fp16 = slice_by_index(begin = var_2184_begin_0, end = var_2184_end_0, end_mask = var_2184_end_mask_0, x = query_5_cast_fp16)[name = string("op_2184_cast_fp16")]; tensor var_2188_begin_0 = const()[name = string("op_2188_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2188_end_0 = const()[name = string("op_2188_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_2188_end_mask_0 = const()[name = string("op_2188_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2188_cast_fp16 = slice_by_index(begin = var_2188_begin_0, end = var_2188_end_0, end_mask = var_2188_end_mask_0, x = query_5_cast_fp16)[name = string("op_2188_cast_fp16")]; tensor var_2192_begin_0 = const()[name = string("op_2192_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_2192_end_0 = const()[name = string("op_2192_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_2192_end_mask_0 = const()[name = string("op_2192_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2192_cast_fp16 = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, x = query_5_cast_fp16)[name = string("op_2192_cast_fp16")]; tensor var_2201_begin_0 = const()[name = string("op_2201_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2201_end_0 = const()[name = string("op_2201_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2201_end_mask_0 = const()[name = string("op_2201_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2201_cast_fp16 = slice_by_index(begin = var_2201_begin_0, end = var_2201_end_0, end_mask = var_2201_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2201_cast_fp16")]; tensor var_2208_begin_0 = const()[name = string("op_2208_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2208_end_0 = const()[name = string("op_2208_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2208_end_mask_0 = const()[name = string("op_2208_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2208_cast_fp16 = slice_by_index(begin = var_2208_begin_0, end = var_2208_end_0, end_mask = var_2208_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2208_cast_fp16")]; tensor var_2215_begin_0 = const()[name = string("op_2215_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2215_end_0 = const()[name = string("op_2215_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2215_end_mask_0 = const()[name = string("op_2215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2215_cast_fp16 = slice_by_index(begin = var_2215_begin_0, end = var_2215_end_0, end_mask = var_2215_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2215_cast_fp16")]; tensor var_2222_begin_0 = const()[name = string("op_2222_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2222_end_0 = const()[name = string("op_2222_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2222_end_mask_0 = const()[name = string("op_2222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2222_cast_fp16 = slice_by_index(begin = var_2222_begin_0, end = var_2222_end_0, end_mask = var_2222_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2222_cast_fp16")]; tensor var_2229_begin_0 = const()[name = string("op_2229_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2229_end_0 = const()[name = string("op_2229_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2229_end_mask_0 = const()[name = string("op_2229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2229_cast_fp16 = slice_by_index(begin = var_2229_begin_0, end = var_2229_end_0, end_mask = var_2229_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2229_cast_fp16")]; tensor var_2236_begin_0 = const()[name = string("op_2236_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2236_end_0 = const()[name = string("op_2236_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2236_end_mask_0 = const()[name = string("op_2236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2236_cast_fp16 = slice_by_index(begin = var_2236_begin_0, end = var_2236_end_0, end_mask = var_2236_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2236_cast_fp16")]; tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2243_cast_fp16")]; tensor var_2250_begin_0 = const()[name = string("op_2250_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2250_end_0 = const()[name = string("op_2250_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2250_end_mask_0 = const()[name = string("op_2250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2250_cast_fp16 = slice_by_index(begin = var_2250_begin_0, end = var_2250_end_0, end_mask = var_2250_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2250_cast_fp16")]; tensor var_2257_begin_0 = const()[name = string("op_2257_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2257_end_0 = const()[name = string("op_2257_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2257_end_mask_0 = const()[name = string("op_2257_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2257_cast_fp16 = slice_by_index(begin = var_2257_begin_0, end = var_2257_end_0, end_mask = var_2257_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2257_cast_fp16")]; tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2264_cast_fp16")]; tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2271_cast_fp16")]; tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2278_cast_fp16")]; tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2285_cast_fp16")]; tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2292_cast_fp16")]; tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2299_cast_fp16")]; tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2306_cast_fp16")]; tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2313_cast_fp16")]; tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2320_cast_fp16")]; tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2327_cast_fp16")]; tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2334_cast_fp16")]; tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2341_cast_fp16")]; tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2348_cast_fp16")]; tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2355_cast_fp16")]; tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2362_cast_fp16")]; tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2369_cast_fp16")]; tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2376_cast_fp16")]; tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2383_cast_fp16")]; tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2390_cast_fp16")]; tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2397_cast_fp16")]; tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2404_cast_fp16")]; tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2411_cast_fp16")]; tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2418_cast_fp16")]; tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2425_cast_fp16")]; tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2432_cast_fp16")]; tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2439_cast_fp16")]; tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2446_cast_fp16")]; tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2453_cast_fp16")]; tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2460_cast_fp16")]; tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2467_cast_fp16")]; tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2474_cast_fp16")]; tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2481_cast_fp16")]; tensor var_2488_begin_0 = const()[name = string("op_2488_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2488_end_0 = const()[name = string("op_2488_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2488_end_mask_0 = const()[name = string("op_2488_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2488_cast_fp16 = slice_by_index(begin = var_2488_begin_0, end = var_2488_end_0, end_mask = var_2488_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2488_cast_fp16")]; tensor var_2495_begin_0 = const()[name = string("op_2495_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2495_end_0 = const()[name = string("op_2495_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2495_end_mask_0 = const()[name = string("op_2495_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2495_cast_fp16 = slice_by_index(begin = var_2495_begin_0, end = var_2495_end_0, end_mask = var_2495_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2495_cast_fp16")]; tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2502_cast_fp16")]; tensor var_2509_begin_0 = const()[name = string("op_2509_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2509_end_0 = const()[name = string("op_2509_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2509_end_mask_0 = const()[name = string("op_2509_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2509_cast_fp16 = slice_by_index(begin = var_2509_begin_0, end = var_2509_end_0, end_mask = var_2509_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2509_cast_fp16")]; tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2516_cast_fp16")]; tensor var_2523_begin_0 = const()[name = string("op_2523_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2523_end_0 = const()[name = string("op_2523_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2523_end_mask_0 = const()[name = string("op_2523_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2523_cast_fp16 = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2523_cast_fp16")]; tensor var_2530_begin_0 = const()[name = string("op_2530_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2530_end_0 = const()[name = string("op_2530_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2530_end_mask_0 = const()[name = string("op_2530_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2530_cast_fp16")]; tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_2535_begin_0 = const()[name = string("op_2535_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2535_end_0 = const()[name = string("op_2535_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_2535_end_mask_0 = const()[name = string("op_2535_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_9")]; tensor var_2535_cast_fp16 = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = k_5_cast_fp16)[name = string("op_2535_cast_fp16")]; tensor var_2539_begin_0 = const()[name = string("op_2539_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_2539_end_0 = const()[name = string("op_2539_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_2539_end_mask_0 = const()[name = string("op_2539_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2539_cast_fp16 = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = k_5_cast_fp16)[name = string("op_2539_cast_fp16")]; tensor var_2543_begin_0 = const()[name = string("op_2543_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_2543_end_0 = const()[name = string("op_2543_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_2543_end_mask_0 = const()[name = string("op_2543_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = k_5_cast_fp16)[name = string("op_2543_cast_fp16")]; tensor var_2547_begin_0 = const()[name = string("op_2547_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_2547_end_0 = const()[name = string("op_2547_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_2547_end_mask_0 = const()[name = string("op_2547_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2547_cast_fp16 = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = k_5_cast_fp16)[name = string("op_2547_cast_fp16")]; tensor var_2551_begin_0 = const()[name = string("op_2551_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_2551_end_0 = const()[name = string("op_2551_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_2551_end_mask_0 = const()[name = string("op_2551_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2551_cast_fp16 = slice_by_index(begin = var_2551_begin_0, end = var_2551_end_0, end_mask = var_2551_end_mask_0, x = k_5_cast_fp16)[name = string("op_2551_cast_fp16")]; tensor var_2555_begin_0 = const()[name = string("op_2555_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_2555_end_0 = const()[name = string("op_2555_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_2555_end_mask_0 = const()[name = string("op_2555_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2555_cast_fp16 = slice_by_index(begin = var_2555_begin_0, end = var_2555_end_0, end_mask = var_2555_end_mask_0, x = k_5_cast_fp16)[name = string("op_2555_cast_fp16")]; tensor var_2559_begin_0 = const()[name = string("op_2559_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_2559_end_0 = const()[name = string("op_2559_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_2559_end_mask_0 = const()[name = string("op_2559_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2559_cast_fp16 = slice_by_index(begin = var_2559_begin_0, end = var_2559_end_0, end_mask = var_2559_end_mask_0, x = k_5_cast_fp16)[name = string("op_2559_cast_fp16")]; tensor var_2563_begin_0 = const()[name = string("op_2563_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_2563_end_0 = const()[name = string("op_2563_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_2563_end_mask_0 = const()[name = string("op_2563_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2563_cast_fp16 = slice_by_index(begin = var_2563_begin_0, end = var_2563_end_0, end_mask = var_2563_end_mask_0, x = k_5_cast_fp16)[name = string("op_2563_cast_fp16")]; tensor var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = k_5_cast_fp16)[name = string("op_2567_cast_fp16")]; tensor var_2571_begin_0 = const()[name = string("op_2571_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_2571_end_0 = const()[name = string("op_2571_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_2571_end_mask_0 = const()[name = string("op_2571_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2571_cast_fp16 = slice_by_index(begin = var_2571_begin_0, end = var_2571_end_0, end_mask = var_2571_end_mask_0, x = k_5_cast_fp16)[name = string("op_2571_cast_fp16")]; tensor var_2575_begin_0 = const()[name = string("op_2575_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_2575_end_0 = const()[name = string("op_2575_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_2575_end_mask_0 = const()[name = string("op_2575_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2575_cast_fp16 = slice_by_index(begin = var_2575_begin_0, end = var_2575_end_0, end_mask = var_2575_end_mask_0, x = k_5_cast_fp16)[name = string("op_2575_cast_fp16")]; tensor var_2579_begin_0 = const()[name = string("op_2579_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_2579_end_0 = const()[name = string("op_2579_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_2579_end_mask_0 = const()[name = string("op_2579_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2579_cast_fp16 = slice_by_index(begin = var_2579_begin_0, end = var_2579_end_0, end_mask = var_2579_end_mask_0, x = k_5_cast_fp16)[name = string("op_2579_cast_fp16")]; tensor var_2581_begin_0 = const()[name = string("op_2581_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2581_end_0 = const()[name = string("op_2581_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2581_end_mask_0 = const()[name = string("op_2581_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2581_cast_fp16 = slice_by_index(begin = var_2581_begin_0, end = var_2581_end_0, end_mask = var_2581_end_mask_0, x = value_5_cast_fp16)[name = string("op_2581_cast_fp16")]; tensor var_2585_begin_0 = const()[name = string("op_2585_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_2585_end_0 = const()[name = string("op_2585_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_2585_end_mask_0 = const()[name = string("op_2585_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2585_cast_fp16 = slice_by_index(begin = var_2585_begin_0, end = var_2585_end_0, end_mask = var_2585_end_mask_0, x = value_5_cast_fp16)[name = string("op_2585_cast_fp16")]; tensor var_2589_begin_0 = const()[name = string("op_2589_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_2589_end_0 = const()[name = string("op_2589_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_2589_end_mask_0 = const()[name = string("op_2589_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2589_cast_fp16 = slice_by_index(begin = var_2589_begin_0, end = var_2589_end_0, end_mask = var_2589_end_mask_0, x = value_5_cast_fp16)[name = string("op_2589_cast_fp16")]; tensor var_2593_begin_0 = const()[name = string("op_2593_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_2593_end_0 = const()[name = string("op_2593_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_2593_end_mask_0 = const()[name = string("op_2593_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2593_cast_fp16 = slice_by_index(begin = var_2593_begin_0, end = var_2593_end_0, end_mask = var_2593_end_mask_0, x = value_5_cast_fp16)[name = string("op_2593_cast_fp16")]; tensor var_2597_begin_0 = const()[name = string("op_2597_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_2597_end_0 = const()[name = string("op_2597_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_2597_end_mask_0 = const()[name = string("op_2597_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2597_cast_fp16 = slice_by_index(begin = var_2597_begin_0, end = var_2597_end_0, end_mask = var_2597_end_mask_0, x = value_5_cast_fp16)[name = string("op_2597_cast_fp16")]; tensor var_2601_begin_0 = const()[name = string("op_2601_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2601_end_0 = const()[name = string("op_2601_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_2601_end_mask_0 = const()[name = string("op_2601_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2601_cast_fp16 = slice_by_index(begin = var_2601_begin_0, end = var_2601_end_0, end_mask = var_2601_end_mask_0, x = value_5_cast_fp16)[name = string("op_2601_cast_fp16")]; tensor var_2605_begin_0 = const()[name = string("op_2605_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_2605_end_0 = const()[name = string("op_2605_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_2605_end_mask_0 = const()[name = string("op_2605_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2605_cast_fp16 = slice_by_index(begin = var_2605_begin_0, end = var_2605_end_0, end_mask = var_2605_end_mask_0, x = value_5_cast_fp16)[name = string("op_2605_cast_fp16")]; tensor var_2609_begin_0 = const()[name = string("op_2609_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_2609_end_0 = const()[name = string("op_2609_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_2609_end_mask_0 = const()[name = string("op_2609_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2609_cast_fp16 = slice_by_index(begin = var_2609_begin_0, end = var_2609_end_0, end_mask = var_2609_end_mask_0, x = value_5_cast_fp16)[name = string("op_2609_cast_fp16")]; tensor var_2613_begin_0 = const()[name = string("op_2613_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_2613_end_0 = const()[name = string("op_2613_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_2613_end_mask_0 = const()[name = string("op_2613_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2613_cast_fp16 = slice_by_index(begin = var_2613_begin_0, end = var_2613_end_0, end_mask = var_2613_end_mask_0, x = value_5_cast_fp16)[name = string("op_2613_cast_fp16")]; tensor var_2617_begin_0 = const()[name = string("op_2617_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_2617_end_0 = const()[name = string("op_2617_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_2617_end_mask_0 = const()[name = string("op_2617_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2617_cast_fp16 = slice_by_index(begin = var_2617_begin_0, end = var_2617_end_0, end_mask = var_2617_end_mask_0, x = value_5_cast_fp16)[name = string("op_2617_cast_fp16")]; tensor var_2621_begin_0 = const()[name = string("op_2621_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2621_end_0 = const()[name = string("op_2621_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_2621_end_mask_0 = const()[name = string("op_2621_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2621_cast_fp16 = slice_by_index(begin = var_2621_begin_0, end = var_2621_end_0, end_mask = var_2621_end_mask_0, x = value_5_cast_fp16)[name = string("op_2621_cast_fp16")]; tensor var_2625_begin_0 = const()[name = string("op_2625_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_2625_end_0 = const()[name = string("op_2625_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_2625_end_mask_0 = const()[name = string("op_2625_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2625_cast_fp16 = slice_by_index(begin = var_2625_begin_0, end = var_2625_end_0, end_mask = var_2625_end_mask_0, x = value_5_cast_fp16)[name = string("op_2625_cast_fp16")]; string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2535_cast_fp16, var_2201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2535_cast_fp16, var_2208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2535_cast_fp16, var_2215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2535_cast_fp16, var_2222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2539_cast_fp16, var_2229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2539_cast_fp16, var_2236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2539_cast_fp16, var_2243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2539_cast_fp16, var_2250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2543_cast_fp16, var_2257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2543_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2543_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2543_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2547_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2547_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2547_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2547_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2551_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2551_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2551_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2551_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2555_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2555_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2555_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2555_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2559_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2559_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2559_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2559_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2563_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2563_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2563_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2563_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_2567_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_2567_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_2567_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_2567_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_2571_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_2571_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_2571_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_2571_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_2575_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_2575_cast_fp16, var_2488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_2575_cast_fp16, var_2495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_2575_cast_fp16, var_2502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_2579_cast_fp16, var_2509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_2579_cast_fp16, var_2516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_2579_cast_fp16, var_2523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_2579_cast_fp16, var_2530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; fp16 var_2724_to_fp16 = const()[name = string("op_2724_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2724_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; fp16 var_2726_to_fp16 = const()[name = string("op_2726_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2726_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; fp16 var_2728_to_fp16 = const()[name = string("op_2728_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2728_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; fp16 var_2730_to_fp16 = const()[name = string("op_2730_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2730_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; fp16 var_2732_to_fp16 = const()[name = string("op_2732_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2732_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2734_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; fp16 var_2736_to_fp16 = const()[name = string("op_2736_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2736_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; fp16 var_2738_to_fp16 = const()[name = string("op_2738_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2738_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; fp16 var_2740_to_fp16 = const()[name = string("op_2740_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2740_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; fp16 var_2742_to_fp16 = const()[name = string("op_2742_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2742_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; fp16 var_2744_to_fp16 = const()[name = string("op_2744_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2744_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; fp16 var_2746_to_fp16 = const()[name = string("op_2746_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2746_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; fp16 var_2748_to_fp16 = const()[name = string("op_2748_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2748_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; fp16 var_2750_to_fp16 = const()[name = string("op_2750_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2750_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; fp16 var_2752_to_fp16 = const()[name = string("op_2752_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2752_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; fp16 var_2754_to_fp16 = const()[name = string("op_2754_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2754_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; fp16 var_2756_to_fp16 = const()[name = string("op_2756_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2756_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; fp16 var_2758_to_fp16 = const()[name = string("op_2758_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2758_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; fp16 var_2760_to_fp16 = const()[name = string("op_2760_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2760_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; fp16 var_2762_to_fp16 = const()[name = string("op_2762_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2762_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; fp16 var_2764_to_fp16 = const()[name = string("op_2764_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2764_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; fp16 var_2766_to_fp16 = const()[name = string("op_2766_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2766_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; fp16 var_2768_to_fp16 = const()[name = string("op_2768_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2768_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; fp16 var_2770_to_fp16 = const()[name = string("op_2770_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2770_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; fp16 var_2772_to_fp16 = const()[name = string("op_2772_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2772_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; fp16 var_2774_to_fp16 = const()[name = string("op_2774_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2774_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; fp16 var_2776_to_fp16 = const()[name = string("op_2776_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2776_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; fp16 var_2778_to_fp16 = const()[name = string("op_2778_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2778_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; fp16 var_2780_to_fp16 = const()[name = string("op_2780_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2780_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; fp16 var_2782_to_fp16 = const()[name = string("op_2782_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2782_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; fp16 var_2784_to_fp16 = const()[name = string("op_2784_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2784_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; fp16 var_2786_to_fp16 = const()[name = string("op_2786_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2786_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; fp16 var_2788_to_fp16 = const()[name = string("op_2788_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_2788_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; fp16 var_2790_to_fp16 = const()[name = string("op_2790_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_2790_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; fp16 var_2792_to_fp16 = const()[name = string("op_2792_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_2792_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; fp16 var_2794_to_fp16 = const()[name = string("op_2794_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_2794_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; fp16 var_2796_to_fp16 = const()[name = string("op_2796_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_2796_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; fp16 var_2798_to_fp16 = const()[name = string("op_2798_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_2798_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; fp16 var_2800_to_fp16 = const()[name = string("op_2800_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_2800_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; fp16 var_2802_to_fp16 = const()[name = string("op_2802_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_2802_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; fp16 var_2804_to_fp16 = const()[name = string("op_2804_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_2804_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; fp16 var_2806_to_fp16 = const()[name = string("op_2806_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_2806_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; fp16 var_2808_to_fp16 = const()[name = string("op_2808_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_2808_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; fp16 var_2810_to_fp16 = const()[name = string("op_2810_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_2810_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; fp16 var_2812_to_fp16 = const()[name = string("op_2812_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_2812_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; fp16 var_2814_to_fp16 = const()[name = string("op_2814_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_2814_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; fp16 var_2816_to_fp16 = const()[name = string("op_2816_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_2816_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; fp16 var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_2818_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; tensor var_2820_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_193_cast_fp16)[name = string("op_2820_cast_fp16")]; tensor var_2821_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_195_cast_fp16)[name = string("op_2821_cast_fp16")]; tensor var_2822_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_197_cast_fp16)[name = string("op_2822_cast_fp16")]; tensor var_2823_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_199_cast_fp16)[name = string("op_2823_cast_fp16")]; tensor var_2824_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_201_cast_fp16)[name = string("op_2824_cast_fp16")]; tensor var_2825_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_203_cast_fp16)[name = string("op_2825_cast_fp16")]; tensor var_2826_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_205_cast_fp16)[name = string("op_2826_cast_fp16")]; tensor var_2827_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_207_cast_fp16)[name = string("op_2827_cast_fp16")]; tensor var_2828_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_209_cast_fp16)[name = string("op_2828_cast_fp16")]; tensor var_2829_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_211_cast_fp16)[name = string("op_2829_cast_fp16")]; tensor var_2830_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_213_cast_fp16)[name = string("op_2830_cast_fp16")]; tensor var_2831_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_215_cast_fp16)[name = string("op_2831_cast_fp16")]; tensor var_2832_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_217_cast_fp16)[name = string("op_2832_cast_fp16")]; tensor var_2833_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_219_cast_fp16)[name = string("op_2833_cast_fp16")]; tensor var_2834_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_221_cast_fp16)[name = string("op_2834_cast_fp16")]; tensor var_2835_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_223_cast_fp16)[name = string("op_2835_cast_fp16")]; tensor var_2836_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_225_cast_fp16)[name = string("op_2836_cast_fp16")]; tensor var_2837_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_227_cast_fp16)[name = string("op_2837_cast_fp16")]; tensor var_2838_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_229_cast_fp16)[name = string("op_2838_cast_fp16")]; tensor var_2839_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_231_cast_fp16)[name = string("op_2839_cast_fp16")]; tensor var_2840_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_233_cast_fp16)[name = string("op_2840_cast_fp16")]; tensor var_2841_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_235_cast_fp16)[name = string("op_2841_cast_fp16")]; tensor var_2842_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_237_cast_fp16)[name = string("op_2842_cast_fp16")]; tensor var_2843_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_239_cast_fp16)[name = string("op_2843_cast_fp16")]; tensor var_2844_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_241_cast_fp16)[name = string("op_2844_cast_fp16")]; tensor var_2845_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_243_cast_fp16)[name = string("op_2845_cast_fp16")]; tensor var_2846_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_245_cast_fp16)[name = string("op_2846_cast_fp16")]; tensor var_2847_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_247_cast_fp16)[name = string("op_2847_cast_fp16")]; tensor var_2848_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_249_cast_fp16)[name = string("op_2848_cast_fp16")]; tensor var_2849_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_251_cast_fp16)[name = string("op_2849_cast_fp16")]; tensor var_2850_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_253_cast_fp16)[name = string("op_2850_cast_fp16")]; tensor var_2851_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_255_cast_fp16)[name = string("op_2851_cast_fp16")]; tensor var_2852_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_257_cast_fp16)[name = string("op_2852_cast_fp16")]; tensor var_2853_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_259_cast_fp16)[name = string("op_2853_cast_fp16")]; tensor var_2854_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_261_cast_fp16)[name = string("op_2854_cast_fp16")]; tensor var_2855_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_263_cast_fp16)[name = string("op_2855_cast_fp16")]; tensor var_2856_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_265_cast_fp16)[name = string("op_2856_cast_fp16")]; tensor var_2857_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_267_cast_fp16)[name = string("op_2857_cast_fp16")]; tensor var_2858_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_269_cast_fp16)[name = string("op_2858_cast_fp16")]; tensor var_2859_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_271_cast_fp16)[name = string("op_2859_cast_fp16")]; tensor var_2860_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_273_cast_fp16)[name = string("op_2860_cast_fp16")]; tensor var_2861_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_275_cast_fp16)[name = string("op_2861_cast_fp16")]; tensor var_2862_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_277_cast_fp16)[name = string("op_2862_cast_fp16")]; tensor var_2863_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_279_cast_fp16)[name = string("op_2863_cast_fp16")]; tensor var_2864_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_281_cast_fp16)[name = string("op_2864_cast_fp16")]; tensor var_2865_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_283_cast_fp16)[name = string("op_2865_cast_fp16")]; tensor var_2866_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_285_cast_fp16)[name = string("op_2866_cast_fp16")]; tensor var_2867_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_287_cast_fp16)[name = string("op_2867_cast_fp16")]; string var_2869_equation_0 = const()[name = string("op_2869_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2869_cast_fp16 = einsum(equation = var_2869_equation_0, values = (var_2581_cast_fp16, var_2820_cast_fp16))[name = string("op_2869_cast_fp16")]; string var_2871_equation_0 = const()[name = string("op_2871_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2871_cast_fp16 = einsum(equation = var_2871_equation_0, values = (var_2581_cast_fp16, var_2821_cast_fp16))[name = string("op_2871_cast_fp16")]; string var_2873_equation_0 = const()[name = string("op_2873_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2873_cast_fp16 = einsum(equation = var_2873_equation_0, values = (var_2581_cast_fp16, var_2822_cast_fp16))[name = string("op_2873_cast_fp16")]; string var_2875_equation_0 = const()[name = string("op_2875_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2875_cast_fp16 = einsum(equation = var_2875_equation_0, values = (var_2581_cast_fp16, var_2823_cast_fp16))[name = string("op_2875_cast_fp16")]; string var_2877_equation_0 = const()[name = string("op_2877_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2877_cast_fp16 = einsum(equation = var_2877_equation_0, values = (var_2585_cast_fp16, var_2824_cast_fp16))[name = string("op_2877_cast_fp16")]; string var_2879_equation_0 = const()[name = string("op_2879_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2879_cast_fp16 = einsum(equation = var_2879_equation_0, values = (var_2585_cast_fp16, var_2825_cast_fp16))[name = string("op_2879_cast_fp16")]; string var_2881_equation_0 = const()[name = string("op_2881_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2881_cast_fp16 = einsum(equation = var_2881_equation_0, values = (var_2585_cast_fp16, var_2826_cast_fp16))[name = string("op_2881_cast_fp16")]; string var_2883_equation_0 = const()[name = string("op_2883_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2883_cast_fp16 = einsum(equation = var_2883_equation_0, values = (var_2585_cast_fp16, var_2827_cast_fp16))[name = string("op_2883_cast_fp16")]; string var_2885_equation_0 = const()[name = string("op_2885_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2885_cast_fp16 = einsum(equation = var_2885_equation_0, values = (var_2589_cast_fp16, var_2828_cast_fp16))[name = string("op_2885_cast_fp16")]; string var_2887_equation_0 = const()[name = string("op_2887_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2887_cast_fp16 = einsum(equation = var_2887_equation_0, values = (var_2589_cast_fp16, var_2829_cast_fp16))[name = string("op_2887_cast_fp16")]; string var_2889_equation_0 = const()[name = string("op_2889_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2889_cast_fp16 = einsum(equation = var_2889_equation_0, values = (var_2589_cast_fp16, var_2830_cast_fp16))[name = string("op_2889_cast_fp16")]; string var_2891_equation_0 = const()[name = string("op_2891_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2891_cast_fp16 = einsum(equation = var_2891_equation_0, values = (var_2589_cast_fp16, var_2831_cast_fp16))[name = string("op_2891_cast_fp16")]; string var_2893_equation_0 = const()[name = string("op_2893_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2893_cast_fp16 = einsum(equation = var_2893_equation_0, values = (var_2593_cast_fp16, var_2832_cast_fp16))[name = string("op_2893_cast_fp16")]; string var_2895_equation_0 = const()[name = string("op_2895_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2895_cast_fp16 = einsum(equation = var_2895_equation_0, values = (var_2593_cast_fp16, var_2833_cast_fp16))[name = string("op_2895_cast_fp16")]; string var_2897_equation_0 = const()[name = string("op_2897_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2897_cast_fp16 = einsum(equation = var_2897_equation_0, values = (var_2593_cast_fp16, var_2834_cast_fp16))[name = string("op_2897_cast_fp16")]; string var_2899_equation_0 = const()[name = string("op_2899_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2899_cast_fp16 = einsum(equation = var_2899_equation_0, values = (var_2593_cast_fp16, var_2835_cast_fp16))[name = string("op_2899_cast_fp16")]; string var_2901_equation_0 = const()[name = string("op_2901_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2901_cast_fp16 = einsum(equation = var_2901_equation_0, values = (var_2597_cast_fp16, var_2836_cast_fp16))[name = string("op_2901_cast_fp16")]; string var_2903_equation_0 = const()[name = string("op_2903_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2903_cast_fp16 = einsum(equation = var_2903_equation_0, values = (var_2597_cast_fp16, var_2837_cast_fp16))[name = string("op_2903_cast_fp16")]; string var_2905_equation_0 = const()[name = string("op_2905_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2905_cast_fp16 = einsum(equation = var_2905_equation_0, values = (var_2597_cast_fp16, var_2838_cast_fp16))[name = string("op_2905_cast_fp16")]; string var_2907_equation_0 = const()[name = string("op_2907_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2597_cast_fp16, var_2839_cast_fp16))[name = string("op_2907_cast_fp16")]; string var_2909_equation_0 = const()[name = string("op_2909_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2601_cast_fp16, var_2840_cast_fp16))[name = string("op_2909_cast_fp16")]; string var_2911_equation_0 = const()[name = string("op_2911_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2601_cast_fp16, var_2841_cast_fp16))[name = string("op_2911_cast_fp16")]; string var_2913_equation_0 = const()[name = string("op_2913_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2601_cast_fp16, var_2842_cast_fp16))[name = string("op_2913_cast_fp16")]; string var_2915_equation_0 = const()[name = string("op_2915_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2601_cast_fp16, var_2843_cast_fp16))[name = string("op_2915_cast_fp16")]; string var_2917_equation_0 = const()[name = string("op_2917_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2605_cast_fp16, var_2844_cast_fp16))[name = string("op_2917_cast_fp16")]; string var_2919_equation_0 = const()[name = string("op_2919_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2605_cast_fp16, var_2845_cast_fp16))[name = string("op_2919_cast_fp16")]; string var_2921_equation_0 = const()[name = string("op_2921_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2605_cast_fp16, var_2846_cast_fp16))[name = string("op_2921_cast_fp16")]; string var_2923_equation_0 = const()[name = string("op_2923_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2605_cast_fp16, var_2847_cast_fp16))[name = string("op_2923_cast_fp16")]; string var_2925_equation_0 = const()[name = string("op_2925_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2609_cast_fp16, var_2848_cast_fp16))[name = string("op_2925_cast_fp16")]; string var_2927_equation_0 = const()[name = string("op_2927_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2609_cast_fp16, var_2849_cast_fp16))[name = string("op_2927_cast_fp16")]; string var_2929_equation_0 = const()[name = string("op_2929_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2609_cast_fp16, var_2850_cast_fp16))[name = string("op_2929_cast_fp16")]; string var_2931_equation_0 = const()[name = string("op_2931_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2609_cast_fp16, var_2851_cast_fp16))[name = string("op_2931_cast_fp16")]; string var_2933_equation_0 = const()[name = string("op_2933_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2613_cast_fp16, var_2852_cast_fp16))[name = string("op_2933_cast_fp16")]; string var_2935_equation_0 = const()[name = string("op_2935_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2613_cast_fp16, var_2853_cast_fp16))[name = string("op_2935_cast_fp16")]; string var_2937_equation_0 = const()[name = string("op_2937_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2613_cast_fp16, var_2854_cast_fp16))[name = string("op_2937_cast_fp16")]; string var_2939_equation_0 = const()[name = string("op_2939_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2939_cast_fp16 = einsum(equation = var_2939_equation_0, values = (var_2613_cast_fp16, var_2855_cast_fp16))[name = string("op_2939_cast_fp16")]; string var_2941_equation_0 = const()[name = string("op_2941_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2941_cast_fp16 = einsum(equation = var_2941_equation_0, values = (var_2617_cast_fp16, var_2856_cast_fp16))[name = string("op_2941_cast_fp16")]; string var_2943_equation_0 = const()[name = string("op_2943_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2943_cast_fp16 = einsum(equation = var_2943_equation_0, values = (var_2617_cast_fp16, var_2857_cast_fp16))[name = string("op_2943_cast_fp16")]; string var_2945_equation_0 = const()[name = string("op_2945_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2945_cast_fp16 = einsum(equation = var_2945_equation_0, values = (var_2617_cast_fp16, var_2858_cast_fp16))[name = string("op_2945_cast_fp16")]; string var_2947_equation_0 = const()[name = string("op_2947_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2947_cast_fp16 = einsum(equation = var_2947_equation_0, values = (var_2617_cast_fp16, var_2859_cast_fp16))[name = string("op_2947_cast_fp16")]; string var_2949_equation_0 = const()[name = string("op_2949_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2949_cast_fp16 = einsum(equation = var_2949_equation_0, values = (var_2621_cast_fp16, var_2860_cast_fp16))[name = string("op_2949_cast_fp16")]; string var_2951_equation_0 = const()[name = string("op_2951_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2951_cast_fp16 = einsum(equation = var_2951_equation_0, values = (var_2621_cast_fp16, var_2861_cast_fp16))[name = string("op_2951_cast_fp16")]; string var_2953_equation_0 = const()[name = string("op_2953_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2953_cast_fp16 = einsum(equation = var_2953_equation_0, values = (var_2621_cast_fp16, var_2862_cast_fp16))[name = string("op_2953_cast_fp16")]; string var_2955_equation_0 = const()[name = string("op_2955_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2955_cast_fp16 = einsum(equation = var_2955_equation_0, values = (var_2621_cast_fp16, var_2863_cast_fp16))[name = string("op_2955_cast_fp16")]; string var_2957_equation_0 = const()[name = string("op_2957_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2957_cast_fp16 = einsum(equation = var_2957_equation_0, values = (var_2625_cast_fp16, var_2864_cast_fp16))[name = string("op_2957_cast_fp16")]; string var_2959_equation_0 = const()[name = string("op_2959_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2959_cast_fp16 = einsum(equation = var_2959_equation_0, values = (var_2625_cast_fp16, var_2865_cast_fp16))[name = string("op_2959_cast_fp16")]; string var_2961_equation_0 = const()[name = string("op_2961_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2961_cast_fp16 = einsum(equation = var_2961_equation_0, values = (var_2625_cast_fp16, var_2866_cast_fp16))[name = string("op_2961_cast_fp16")]; string var_2963_equation_0 = const()[name = string("op_2963_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_2963_cast_fp16 = einsum(equation = var_2963_equation_0, values = (var_2625_cast_fp16, var_2867_cast_fp16))[name = string("op_2963_cast_fp16")]; bool var_2965_interleave_0 = const()[name = string("op_2965_interleave_0"), val = bool(false)]; tensor var_2965_cast_fp16 = concat(axis = var_2076, interleave = var_2965_interleave_0, values = (var_2869_cast_fp16, var_2871_cast_fp16, var_2873_cast_fp16, var_2875_cast_fp16))[name = string("op_2965_cast_fp16")]; bool var_2967_interleave_0 = const()[name = string("op_2967_interleave_0"), val = bool(false)]; tensor var_2967_cast_fp16 = concat(axis = var_2076, interleave = var_2967_interleave_0, values = (var_2877_cast_fp16, var_2879_cast_fp16, var_2881_cast_fp16, var_2883_cast_fp16))[name = string("op_2967_cast_fp16")]; bool var_2969_interleave_0 = const()[name = string("op_2969_interleave_0"), val = bool(false)]; tensor var_2969_cast_fp16 = concat(axis = var_2076, interleave = var_2969_interleave_0, values = (var_2885_cast_fp16, var_2887_cast_fp16, var_2889_cast_fp16, var_2891_cast_fp16))[name = string("op_2969_cast_fp16")]; bool var_2971_interleave_0 = const()[name = string("op_2971_interleave_0"), val = bool(false)]; tensor var_2971_cast_fp16 = concat(axis = var_2076, interleave = var_2971_interleave_0, values = (var_2893_cast_fp16, var_2895_cast_fp16, var_2897_cast_fp16, var_2899_cast_fp16))[name = string("op_2971_cast_fp16")]; bool var_2973_interleave_0 = const()[name = string("op_2973_interleave_0"), val = bool(false)]; tensor var_2973_cast_fp16 = concat(axis = var_2076, interleave = var_2973_interleave_0, values = (var_2901_cast_fp16, var_2903_cast_fp16, var_2905_cast_fp16, var_2907_cast_fp16))[name = string("op_2973_cast_fp16")]; bool var_2975_interleave_0 = const()[name = string("op_2975_interleave_0"), val = bool(false)]; tensor var_2975_cast_fp16 = concat(axis = var_2076, interleave = var_2975_interleave_0, values = (var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16))[name = string("op_2975_cast_fp16")]; bool var_2977_interleave_0 = const()[name = string("op_2977_interleave_0"), val = bool(false)]; tensor var_2977_cast_fp16 = concat(axis = var_2076, interleave = var_2977_interleave_0, values = (var_2917_cast_fp16, var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16))[name = string("op_2977_cast_fp16")]; bool var_2979_interleave_0 = const()[name = string("op_2979_interleave_0"), val = bool(false)]; tensor var_2979_cast_fp16 = concat(axis = var_2076, interleave = var_2979_interleave_0, values = (var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16))[name = string("op_2979_cast_fp16")]; bool var_2981_interleave_0 = const()[name = string("op_2981_interleave_0"), val = bool(false)]; tensor var_2981_cast_fp16 = concat(axis = var_2076, interleave = var_2981_interleave_0, values = (var_2933_cast_fp16, var_2935_cast_fp16, var_2937_cast_fp16, var_2939_cast_fp16))[name = string("op_2981_cast_fp16")]; bool var_2983_interleave_0 = const()[name = string("op_2983_interleave_0"), val = bool(false)]; tensor var_2983_cast_fp16 = concat(axis = var_2076, interleave = var_2983_interleave_0, values = (var_2941_cast_fp16, var_2943_cast_fp16, var_2945_cast_fp16, var_2947_cast_fp16))[name = string("op_2983_cast_fp16")]; bool var_2985_interleave_0 = const()[name = string("op_2985_interleave_0"), val = bool(false)]; tensor var_2985_cast_fp16 = concat(axis = var_2076, interleave = var_2985_interleave_0, values = (var_2949_cast_fp16, var_2951_cast_fp16, var_2953_cast_fp16, var_2955_cast_fp16))[name = string("op_2985_cast_fp16")]; bool var_2987_interleave_0 = const()[name = string("op_2987_interleave_0"), val = bool(false)]; tensor var_2987_cast_fp16 = concat(axis = var_2076, interleave = var_2987_interleave_0, values = (var_2957_cast_fp16, var_2959_cast_fp16, var_2961_cast_fp16, var_2963_cast_fp16))[name = string("op_2987_cast_fp16")]; bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; tensor input_17_cast_fp16 = concat(axis = var_2093, interleave = input_17_interleave_0, values = (var_2965_cast_fp16, var_2967_cast_fp16, var_2969_cast_fp16, var_2971_cast_fp16, var_2973_cast_fp16, var_2975_cast_fp16, var_2977_cast_fp16, var_2979_cast_fp16, var_2981_cast_fp16, var_2983_cast_fp16, var_2985_cast_fp16, var_2987_cast_fp16))[name = string("input_17_cast_fp16")]; string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38114112)))]; tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293824)))]; tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; fp16 var_3006_to_fp16 = const()[name = string("op_3006_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_3006_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39295424)))]; tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297024)))]; fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39298624)))]; tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44017280)))]; tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44023488)))]; tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48742144)))]; tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; int32 var_3035 = const()[name = string("op_3035"), val = int32(3)]; int32 var_3052 = const()[name = string("op_3052"), val = int32(1)]; tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; fp16 var_3069_to_fp16 = const()[name = string("op_3069_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_3069_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48743744)))]; tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48745344)))]; fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48746944)))]; tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49926656)))]; tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49928256)))]; tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51107968)))]; tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52287680)))]; tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = query_7_cast_fp16)[name = string("op_3107_cast_fp16")]; tensor var_3111_begin_0 = const()[name = string("op_3111_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_3111_end_0 = const()[name = string("op_3111_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_3111_end_mask_0 = const()[name = string("op_3111_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3111_cast_fp16 = slice_by_index(begin = var_3111_begin_0, end = var_3111_end_0, end_mask = var_3111_end_mask_0, x = query_7_cast_fp16)[name = string("op_3111_cast_fp16")]; tensor var_3115_begin_0 = const()[name = string("op_3115_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_3115_end_0 = const()[name = string("op_3115_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_3115_end_mask_0 = const()[name = string("op_3115_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3115_cast_fp16 = slice_by_index(begin = var_3115_begin_0, end = var_3115_end_0, end_mask = var_3115_end_mask_0, x = query_7_cast_fp16)[name = string("op_3115_cast_fp16")]; tensor var_3119_begin_0 = const()[name = string("op_3119_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_3119_end_0 = const()[name = string("op_3119_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_3119_end_mask_0 = const()[name = string("op_3119_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3119_cast_fp16 = slice_by_index(begin = var_3119_begin_0, end = var_3119_end_0, end_mask = var_3119_end_mask_0, x = query_7_cast_fp16)[name = string("op_3119_cast_fp16")]; tensor var_3123_begin_0 = const()[name = string("op_3123_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_3123_end_0 = const()[name = string("op_3123_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_3123_end_mask_0 = const()[name = string("op_3123_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3123_cast_fp16 = slice_by_index(begin = var_3123_begin_0, end = var_3123_end_0, end_mask = var_3123_end_mask_0, x = query_7_cast_fp16)[name = string("op_3123_cast_fp16")]; tensor var_3127_begin_0 = const()[name = string("op_3127_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3127_end_0 = const()[name = string("op_3127_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_3127_end_mask_0 = const()[name = string("op_3127_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3127_cast_fp16 = slice_by_index(begin = var_3127_begin_0, end = var_3127_end_0, end_mask = var_3127_end_mask_0, x = query_7_cast_fp16)[name = string("op_3127_cast_fp16")]; tensor var_3131_begin_0 = const()[name = string("op_3131_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_3131_end_0 = const()[name = string("op_3131_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_3131_end_mask_0 = const()[name = string("op_3131_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3131_cast_fp16 = slice_by_index(begin = var_3131_begin_0, end = var_3131_end_0, end_mask = var_3131_end_mask_0, x = query_7_cast_fp16)[name = string("op_3131_cast_fp16")]; tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = query_7_cast_fp16)[name = string("op_3135_cast_fp16")]; tensor var_3139_begin_0 = const()[name = string("op_3139_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_3139_end_0 = const()[name = string("op_3139_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_3139_end_mask_0 = const()[name = string("op_3139_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3139_cast_fp16 = slice_by_index(begin = var_3139_begin_0, end = var_3139_end_0, end_mask = var_3139_end_mask_0, x = query_7_cast_fp16)[name = string("op_3139_cast_fp16")]; tensor var_3143_begin_0 = const()[name = string("op_3143_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_3143_end_0 = const()[name = string("op_3143_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_3143_end_mask_0 = const()[name = string("op_3143_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3143_cast_fp16 = slice_by_index(begin = var_3143_begin_0, end = var_3143_end_0, end_mask = var_3143_end_mask_0, x = query_7_cast_fp16)[name = string("op_3143_cast_fp16")]; tensor var_3147_begin_0 = const()[name = string("op_3147_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3147_end_0 = const()[name = string("op_3147_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_3147_end_mask_0 = const()[name = string("op_3147_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3147_cast_fp16 = slice_by_index(begin = var_3147_begin_0, end = var_3147_end_0, end_mask = var_3147_end_mask_0, x = query_7_cast_fp16)[name = string("op_3147_cast_fp16")]; tensor var_3151_begin_0 = const()[name = string("op_3151_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_3151_end_0 = const()[name = string("op_3151_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_3151_end_mask_0 = const()[name = string("op_3151_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3151_cast_fp16 = slice_by_index(begin = var_3151_begin_0, end = var_3151_end_0, end_mask = var_3151_end_mask_0, x = query_7_cast_fp16)[name = string("op_3151_cast_fp16")]; tensor var_3160_begin_0 = const()[name = string("op_3160_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3160_end_0 = const()[name = string("op_3160_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3160_end_mask_0 = const()[name = string("op_3160_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3160_cast_fp16 = slice_by_index(begin = var_3160_begin_0, end = var_3160_end_0, end_mask = var_3160_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3160_cast_fp16")]; tensor var_3167_begin_0 = const()[name = string("op_3167_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3167_end_0 = const()[name = string("op_3167_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3167_end_mask_0 = const()[name = string("op_3167_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3167_cast_fp16 = slice_by_index(begin = var_3167_begin_0, end = var_3167_end_0, end_mask = var_3167_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3167_cast_fp16")]; tensor var_3174_begin_0 = const()[name = string("op_3174_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3174_end_0 = const()[name = string("op_3174_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3174_end_mask_0 = const()[name = string("op_3174_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3174_cast_fp16 = slice_by_index(begin = var_3174_begin_0, end = var_3174_end_0, end_mask = var_3174_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3174_cast_fp16")]; tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3181_cast_fp16")]; tensor var_3188_begin_0 = const()[name = string("op_3188_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3188_end_0 = const()[name = string("op_3188_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3188_end_mask_0 = const()[name = string("op_3188_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3188_cast_fp16 = slice_by_index(begin = var_3188_begin_0, end = var_3188_end_0, end_mask = var_3188_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3188_cast_fp16")]; tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3195_cast_fp16")]; tensor var_3202_begin_0 = const()[name = string("op_3202_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3202_end_0 = const()[name = string("op_3202_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3202_end_mask_0 = const()[name = string("op_3202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3202_cast_fp16 = slice_by_index(begin = var_3202_begin_0, end = var_3202_end_0, end_mask = var_3202_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3202_cast_fp16")]; tensor var_3209_begin_0 = const()[name = string("op_3209_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3209_end_0 = const()[name = string("op_3209_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3209_end_mask_0 = const()[name = string("op_3209_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3209_cast_fp16 = slice_by_index(begin = var_3209_begin_0, end = var_3209_end_0, end_mask = var_3209_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3209_cast_fp16")]; tensor var_3216_begin_0 = const()[name = string("op_3216_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3216_end_0 = const()[name = string("op_3216_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3216_end_mask_0 = const()[name = string("op_3216_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3216_cast_fp16 = slice_by_index(begin = var_3216_begin_0, end = var_3216_end_0, end_mask = var_3216_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3216_cast_fp16")]; tensor var_3223_begin_0 = const()[name = string("op_3223_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3223_end_0 = const()[name = string("op_3223_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3223_end_mask_0 = const()[name = string("op_3223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3223_cast_fp16 = slice_by_index(begin = var_3223_begin_0, end = var_3223_end_0, end_mask = var_3223_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3223_cast_fp16")]; tensor var_3230_begin_0 = const()[name = string("op_3230_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3230_end_0 = const()[name = string("op_3230_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3230_end_mask_0 = const()[name = string("op_3230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3230_cast_fp16 = slice_by_index(begin = var_3230_begin_0, end = var_3230_end_0, end_mask = var_3230_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3230_cast_fp16")]; tensor var_3237_begin_0 = const()[name = string("op_3237_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3237_end_0 = const()[name = string("op_3237_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3237_end_mask_0 = const()[name = string("op_3237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3237_cast_fp16 = slice_by_index(begin = var_3237_begin_0, end = var_3237_end_0, end_mask = var_3237_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3237_cast_fp16")]; tensor var_3244_begin_0 = const()[name = string("op_3244_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3244_end_0 = const()[name = string("op_3244_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3244_end_mask_0 = const()[name = string("op_3244_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3244_cast_fp16 = slice_by_index(begin = var_3244_begin_0, end = var_3244_end_0, end_mask = var_3244_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3244_cast_fp16")]; tensor var_3251_begin_0 = const()[name = string("op_3251_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3251_end_0 = const()[name = string("op_3251_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3251_end_mask_0 = const()[name = string("op_3251_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3251_cast_fp16 = slice_by_index(begin = var_3251_begin_0, end = var_3251_end_0, end_mask = var_3251_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3251_cast_fp16")]; tensor var_3258_begin_0 = const()[name = string("op_3258_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3258_end_0 = const()[name = string("op_3258_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3258_end_mask_0 = const()[name = string("op_3258_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3258_cast_fp16 = slice_by_index(begin = var_3258_begin_0, end = var_3258_end_0, end_mask = var_3258_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3258_cast_fp16")]; tensor var_3265_begin_0 = const()[name = string("op_3265_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3265_end_0 = const()[name = string("op_3265_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3265_end_mask_0 = const()[name = string("op_3265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3265_cast_fp16 = slice_by_index(begin = var_3265_begin_0, end = var_3265_end_0, end_mask = var_3265_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3265_cast_fp16")]; tensor var_3272_begin_0 = const()[name = string("op_3272_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3272_end_0 = const()[name = string("op_3272_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3272_end_mask_0 = const()[name = string("op_3272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3272_cast_fp16")]; tensor var_3279_begin_0 = const()[name = string("op_3279_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3279_end_0 = const()[name = string("op_3279_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3279_end_mask_0 = const()[name = string("op_3279_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3279_cast_fp16 = slice_by_index(begin = var_3279_begin_0, end = var_3279_end_0, end_mask = var_3279_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3279_cast_fp16")]; tensor var_3286_begin_0 = const()[name = string("op_3286_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3286_end_0 = const()[name = string("op_3286_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3286_end_mask_0 = const()[name = string("op_3286_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3286_cast_fp16 = slice_by_index(begin = var_3286_begin_0, end = var_3286_end_0, end_mask = var_3286_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3286_cast_fp16")]; tensor var_3293_begin_0 = const()[name = string("op_3293_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3293_end_0 = const()[name = string("op_3293_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3293_end_mask_0 = const()[name = string("op_3293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3293_cast_fp16 = slice_by_index(begin = var_3293_begin_0, end = var_3293_end_0, end_mask = var_3293_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3293_cast_fp16")]; tensor var_3300_begin_0 = const()[name = string("op_3300_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3300_end_0 = const()[name = string("op_3300_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3300_end_mask_0 = const()[name = string("op_3300_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3300_cast_fp16 = slice_by_index(begin = var_3300_begin_0, end = var_3300_end_0, end_mask = var_3300_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3300_cast_fp16")]; tensor var_3307_begin_0 = const()[name = string("op_3307_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3307_end_0 = const()[name = string("op_3307_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3307_end_mask_0 = const()[name = string("op_3307_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3307_cast_fp16 = slice_by_index(begin = var_3307_begin_0, end = var_3307_end_0, end_mask = var_3307_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3307_cast_fp16")]; tensor var_3314_begin_0 = const()[name = string("op_3314_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3314_end_0 = const()[name = string("op_3314_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3314_end_mask_0 = const()[name = string("op_3314_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3314_cast_fp16 = slice_by_index(begin = var_3314_begin_0, end = var_3314_end_0, end_mask = var_3314_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3314_cast_fp16")]; tensor var_3321_begin_0 = const()[name = string("op_3321_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3321_end_0 = const()[name = string("op_3321_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3321_end_mask_0 = const()[name = string("op_3321_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3321_cast_fp16 = slice_by_index(begin = var_3321_begin_0, end = var_3321_end_0, end_mask = var_3321_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3321_cast_fp16")]; tensor var_3328_begin_0 = const()[name = string("op_3328_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3328_end_0 = const()[name = string("op_3328_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3328_end_mask_0 = const()[name = string("op_3328_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3328_cast_fp16")]; tensor var_3335_begin_0 = const()[name = string("op_3335_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3335_end_0 = const()[name = string("op_3335_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3335_end_mask_0 = const()[name = string("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3335_cast_fp16")]; tensor var_3342_begin_0 = const()[name = string("op_3342_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3342_end_0 = const()[name = string("op_3342_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3342_end_mask_0 = const()[name = string("op_3342_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3342_cast_fp16 = slice_by_index(begin = var_3342_begin_0, end = var_3342_end_0, end_mask = var_3342_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3342_cast_fp16")]; tensor var_3349_begin_0 = const()[name = string("op_3349_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3349_end_0 = const()[name = string("op_3349_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3349_end_mask_0 = const()[name = string("op_3349_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3349_cast_fp16 = slice_by_index(begin = var_3349_begin_0, end = var_3349_end_0, end_mask = var_3349_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3349_cast_fp16")]; tensor var_3356_begin_0 = const()[name = string("op_3356_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3356_end_0 = const()[name = string("op_3356_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3356_end_mask_0 = const()[name = string("op_3356_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3356_cast_fp16 = slice_by_index(begin = var_3356_begin_0, end = var_3356_end_0, end_mask = var_3356_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3356_cast_fp16")]; tensor var_3363_begin_0 = const()[name = string("op_3363_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3363_end_0 = const()[name = string("op_3363_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3363_end_mask_0 = const()[name = string("op_3363_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3363_cast_fp16 = slice_by_index(begin = var_3363_begin_0, end = var_3363_end_0, end_mask = var_3363_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3363_cast_fp16")]; tensor var_3370_begin_0 = const()[name = string("op_3370_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3370_end_0 = const()[name = string("op_3370_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3370_end_mask_0 = const()[name = string("op_3370_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3370_cast_fp16 = slice_by_index(begin = var_3370_begin_0, end = var_3370_end_0, end_mask = var_3370_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3370_cast_fp16")]; tensor var_3377_begin_0 = const()[name = string("op_3377_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3377_end_0 = const()[name = string("op_3377_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3377_end_mask_0 = const()[name = string("op_3377_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3377_cast_fp16")]; tensor var_3384_begin_0 = const()[name = string("op_3384_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3384_end_0 = const()[name = string("op_3384_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3384_end_mask_0 = const()[name = string("op_3384_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3384_cast_fp16 = slice_by_index(begin = var_3384_begin_0, end = var_3384_end_0, end_mask = var_3384_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3384_cast_fp16")]; tensor var_3391_begin_0 = const()[name = string("op_3391_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3391_end_0 = const()[name = string("op_3391_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3391_end_mask_0 = const()[name = string("op_3391_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3391_cast_fp16 = slice_by_index(begin = var_3391_begin_0, end = var_3391_end_0, end_mask = var_3391_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3391_cast_fp16")]; tensor var_3398_begin_0 = const()[name = string("op_3398_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3398_end_0 = const()[name = string("op_3398_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3398_end_mask_0 = const()[name = string("op_3398_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3398_cast_fp16 = slice_by_index(begin = var_3398_begin_0, end = var_3398_end_0, end_mask = var_3398_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3398_cast_fp16")]; tensor var_3405_begin_0 = const()[name = string("op_3405_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3405_end_0 = const()[name = string("op_3405_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3405_end_mask_0 = const()[name = string("op_3405_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3405_cast_fp16 = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3405_cast_fp16")]; tensor var_3412_begin_0 = const()[name = string("op_3412_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3412_end_0 = const()[name = string("op_3412_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3412_end_mask_0 = const()[name = string("op_3412_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3412_cast_fp16 = slice_by_index(begin = var_3412_begin_0, end = var_3412_end_0, end_mask = var_3412_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3412_cast_fp16")]; tensor var_3419_begin_0 = const()[name = string("op_3419_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3419_end_0 = const()[name = string("op_3419_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3419_end_mask_0 = const()[name = string("op_3419_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3419_cast_fp16 = slice_by_index(begin = var_3419_begin_0, end = var_3419_end_0, end_mask = var_3419_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3419_cast_fp16")]; tensor var_3426_begin_0 = const()[name = string("op_3426_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3426_end_0 = const()[name = string("op_3426_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3426_end_mask_0 = const()[name = string("op_3426_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3426_cast_fp16 = slice_by_index(begin = var_3426_begin_0, end = var_3426_end_0, end_mask = var_3426_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3426_cast_fp16")]; tensor var_3433_begin_0 = const()[name = string("op_3433_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3433_end_0 = const()[name = string("op_3433_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3433_end_mask_0 = const()[name = string("op_3433_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3433_cast_fp16")]; tensor var_3440_begin_0 = const()[name = string("op_3440_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3440_end_0 = const()[name = string("op_3440_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3440_end_mask_0 = const()[name = string("op_3440_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3440_cast_fp16 = slice_by_index(begin = var_3440_begin_0, end = var_3440_end_0, end_mask = var_3440_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3440_cast_fp16")]; tensor var_3447_begin_0 = const()[name = string("op_3447_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3447_end_0 = const()[name = string("op_3447_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3447_end_mask_0 = const()[name = string("op_3447_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3447_cast_fp16 = slice_by_index(begin = var_3447_begin_0, end = var_3447_end_0, end_mask = var_3447_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3447_cast_fp16")]; tensor var_3454_begin_0 = const()[name = string("op_3454_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3454_end_0 = const()[name = string("op_3454_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3454_end_mask_0 = const()[name = string("op_3454_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3454_cast_fp16 = slice_by_index(begin = var_3454_begin_0, end = var_3454_end_0, end_mask = var_3454_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3454_cast_fp16")]; tensor var_3461_begin_0 = const()[name = string("op_3461_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3461_end_0 = const()[name = string("op_3461_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3461_end_mask_0 = const()[name = string("op_3461_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3461_cast_fp16 = slice_by_index(begin = var_3461_begin_0, end = var_3461_end_0, end_mask = var_3461_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3461_cast_fp16")]; tensor var_3468_begin_0 = const()[name = string("op_3468_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3468_end_0 = const()[name = string("op_3468_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3468_end_mask_0 = const()[name = string("op_3468_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3468_cast_fp16 = slice_by_index(begin = var_3468_begin_0, end = var_3468_end_0, end_mask = var_3468_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3468_cast_fp16")]; tensor var_3475_begin_0 = const()[name = string("op_3475_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3475_end_0 = const()[name = string("op_3475_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3475_end_mask_0 = const()[name = string("op_3475_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3475_cast_fp16 = slice_by_index(begin = var_3475_begin_0, end = var_3475_end_0, end_mask = var_3475_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3475_cast_fp16")]; tensor var_3482_begin_0 = const()[name = string("op_3482_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3482_end_0 = const()[name = string("op_3482_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3482_end_mask_0 = const()[name = string("op_3482_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3482_cast_fp16 = slice_by_index(begin = var_3482_begin_0, end = var_3482_end_0, end_mask = var_3482_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3482_cast_fp16")]; tensor var_3489_begin_0 = const()[name = string("op_3489_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3489_end_0 = const()[name = string("op_3489_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3489_end_mask_0 = const()[name = string("op_3489_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3489_cast_fp16 = slice_by_index(begin = var_3489_begin_0, end = var_3489_end_0, end_mask = var_3489_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3489_cast_fp16")]; tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3494_begin_0 = const()[name = string("op_3494_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3494_end_0 = const()[name = string("op_3494_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_3494_end_mask_0 = const()[name = string("op_3494_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_8")]; tensor var_3494_cast_fp16 = slice_by_index(begin = var_3494_begin_0, end = var_3494_end_0, end_mask = var_3494_end_mask_0, x = k_7_cast_fp16)[name = string("op_3494_cast_fp16")]; tensor var_3498_begin_0 = const()[name = string("op_3498_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_3498_end_0 = const()[name = string("op_3498_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_3498_end_mask_0 = const()[name = string("op_3498_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3498_cast_fp16 = slice_by_index(begin = var_3498_begin_0, end = var_3498_end_0, end_mask = var_3498_end_mask_0, x = k_7_cast_fp16)[name = string("op_3498_cast_fp16")]; tensor var_3502_begin_0 = const()[name = string("op_3502_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_3502_end_0 = const()[name = string("op_3502_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_3502_end_mask_0 = const()[name = string("op_3502_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3502_cast_fp16 = slice_by_index(begin = var_3502_begin_0, end = var_3502_end_0, end_mask = var_3502_end_mask_0, x = k_7_cast_fp16)[name = string("op_3502_cast_fp16")]; tensor var_3506_begin_0 = const()[name = string("op_3506_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_3506_end_0 = const()[name = string("op_3506_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_3506_end_mask_0 = const()[name = string("op_3506_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3506_cast_fp16 = slice_by_index(begin = var_3506_begin_0, end = var_3506_end_0, end_mask = var_3506_end_mask_0, x = k_7_cast_fp16)[name = string("op_3506_cast_fp16")]; tensor var_3510_begin_0 = const()[name = string("op_3510_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_3510_end_0 = const()[name = string("op_3510_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_3510_end_mask_0 = const()[name = string("op_3510_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3510_cast_fp16 = slice_by_index(begin = var_3510_begin_0, end = var_3510_end_0, end_mask = var_3510_end_mask_0, x = k_7_cast_fp16)[name = string("op_3510_cast_fp16")]; tensor var_3514_begin_0 = const()[name = string("op_3514_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3514_end_0 = const()[name = string("op_3514_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_3514_end_mask_0 = const()[name = string("op_3514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3514_cast_fp16 = slice_by_index(begin = var_3514_begin_0, end = var_3514_end_0, end_mask = var_3514_end_mask_0, x = k_7_cast_fp16)[name = string("op_3514_cast_fp16")]; tensor var_3518_begin_0 = const()[name = string("op_3518_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_3518_end_0 = const()[name = string("op_3518_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_3518_end_mask_0 = const()[name = string("op_3518_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3518_cast_fp16 = slice_by_index(begin = var_3518_begin_0, end = var_3518_end_0, end_mask = var_3518_end_mask_0, x = k_7_cast_fp16)[name = string("op_3518_cast_fp16")]; tensor var_3522_begin_0 = const()[name = string("op_3522_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_3522_end_0 = const()[name = string("op_3522_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_3522_end_mask_0 = const()[name = string("op_3522_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3522_cast_fp16 = slice_by_index(begin = var_3522_begin_0, end = var_3522_end_0, end_mask = var_3522_end_mask_0, x = k_7_cast_fp16)[name = string("op_3522_cast_fp16")]; tensor var_3526_begin_0 = const()[name = string("op_3526_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_3526_end_0 = const()[name = string("op_3526_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_3526_end_mask_0 = const()[name = string("op_3526_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3526_cast_fp16 = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = k_7_cast_fp16)[name = string("op_3526_cast_fp16")]; tensor var_3530_begin_0 = const()[name = string("op_3530_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_3530_end_0 = const()[name = string("op_3530_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_3530_end_mask_0 = const()[name = string("op_3530_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3530_cast_fp16 = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = k_7_cast_fp16)[name = string("op_3530_cast_fp16")]; tensor var_3534_begin_0 = const()[name = string("op_3534_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3534_end_0 = const()[name = string("op_3534_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_3534_end_mask_0 = const()[name = string("op_3534_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3534_cast_fp16 = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = k_7_cast_fp16)[name = string("op_3534_cast_fp16")]; tensor var_3538_begin_0 = const()[name = string("op_3538_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_3538_end_0 = const()[name = string("op_3538_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_3538_end_mask_0 = const()[name = string("op_3538_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3538_cast_fp16 = slice_by_index(begin = var_3538_begin_0, end = var_3538_end_0, end_mask = var_3538_end_mask_0, x = k_7_cast_fp16)[name = string("op_3538_cast_fp16")]; tensor var_3540_begin_0 = const()[name = string("op_3540_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3540_end_0 = const()[name = string("op_3540_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3540_end_mask_0 = const()[name = string("op_3540_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = value_7_cast_fp16)[name = string("op_3540_cast_fp16")]; tensor var_3544_begin_0 = const()[name = string("op_3544_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_3544_end_0 = const()[name = string("op_3544_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_3544_end_mask_0 = const()[name = string("op_3544_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3544_cast_fp16 = slice_by_index(begin = var_3544_begin_0, end = var_3544_end_0, end_mask = var_3544_end_mask_0, x = value_7_cast_fp16)[name = string("op_3544_cast_fp16")]; tensor var_3548_begin_0 = const()[name = string("op_3548_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_3548_end_0 = const()[name = string("op_3548_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_3548_end_mask_0 = const()[name = string("op_3548_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3548_cast_fp16 = slice_by_index(begin = var_3548_begin_0, end = var_3548_end_0, end_mask = var_3548_end_mask_0, x = value_7_cast_fp16)[name = string("op_3548_cast_fp16")]; tensor var_3552_begin_0 = const()[name = string("op_3552_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_3552_end_0 = const()[name = string("op_3552_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_3552_end_mask_0 = const()[name = string("op_3552_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3552_cast_fp16 = slice_by_index(begin = var_3552_begin_0, end = var_3552_end_0, end_mask = var_3552_end_mask_0, x = value_7_cast_fp16)[name = string("op_3552_cast_fp16")]; tensor var_3556_begin_0 = const()[name = string("op_3556_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_3556_end_0 = const()[name = string("op_3556_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_3556_end_mask_0 = const()[name = string("op_3556_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3556_cast_fp16 = slice_by_index(begin = var_3556_begin_0, end = var_3556_end_0, end_mask = var_3556_end_mask_0, x = value_7_cast_fp16)[name = string("op_3556_cast_fp16")]; tensor var_3560_begin_0 = const()[name = string("op_3560_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3560_end_0 = const()[name = string("op_3560_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_3560_end_mask_0 = const()[name = string("op_3560_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3560_cast_fp16 = slice_by_index(begin = var_3560_begin_0, end = var_3560_end_0, end_mask = var_3560_end_mask_0, x = value_7_cast_fp16)[name = string("op_3560_cast_fp16")]; tensor var_3564_begin_0 = const()[name = string("op_3564_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_3564_end_0 = const()[name = string("op_3564_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_3564_end_mask_0 = const()[name = string("op_3564_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3564_cast_fp16 = slice_by_index(begin = var_3564_begin_0, end = var_3564_end_0, end_mask = var_3564_end_mask_0, x = value_7_cast_fp16)[name = string("op_3564_cast_fp16")]; tensor var_3568_begin_0 = const()[name = string("op_3568_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_3568_end_0 = const()[name = string("op_3568_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_3568_end_mask_0 = const()[name = string("op_3568_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3568_cast_fp16 = slice_by_index(begin = var_3568_begin_0, end = var_3568_end_0, end_mask = var_3568_end_mask_0, x = value_7_cast_fp16)[name = string("op_3568_cast_fp16")]; tensor var_3572_begin_0 = const()[name = string("op_3572_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_3572_end_0 = const()[name = string("op_3572_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_3572_end_mask_0 = const()[name = string("op_3572_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3572_cast_fp16 = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = value_7_cast_fp16)[name = string("op_3572_cast_fp16")]; tensor var_3576_begin_0 = const()[name = string("op_3576_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_3576_end_0 = const()[name = string("op_3576_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_3576_end_mask_0 = const()[name = string("op_3576_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3576_cast_fp16 = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = value_7_cast_fp16)[name = string("op_3576_cast_fp16")]; tensor var_3580_begin_0 = const()[name = string("op_3580_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3580_end_0 = const()[name = string("op_3580_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_3580_end_mask_0 = const()[name = string("op_3580_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3580_cast_fp16 = slice_by_index(begin = var_3580_begin_0, end = var_3580_end_0, end_mask = var_3580_end_mask_0, x = value_7_cast_fp16)[name = string("op_3580_cast_fp16")]; tensor var_3584_begin_0 = const()[name = string("op_3584_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_3584_end_0 = const()[name = string("op_3584_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_3584_end_mask_0 = const()[name = string("op_3584_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3584_cast_fp16 = slice_by_index(begin = var_3584_begin_0, end = var_3584_end_0, end_mask = var_3584_end_mask_0, x = value_7_cast_fp16)[name = string("op_3584_cast_fp16")]; string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3494_cast_fp16, var_3160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3494_cast_fp16, var_3167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3494_cast_fp16, var_3174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3494_cast_fp16, var_3181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3498_cast_fp16, var_3188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3498_cast_fp16, var_3195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3498_cast_fp16, var_3202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3498_cast_fp16, var_3209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3502_cast_fp16, var_3216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3502_cast_fp16, var_3223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3502_cast_fp16, var_3230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3502_cast_fp16, var_3237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3506_cast_fp16, var_3244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3506_cast_fp16, var_3251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3506_cast_fp16, var_3258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3506_cast_fp16, var_3265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3510_cast_fp16, var_3272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3510_cast_fp16, var_3279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3510_cast_fp16, var_3286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3510_cast_fp16, var_3293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3514_cast_fp16, var_3300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3514_cast_fp16, var_3307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3514_cast_fp16, var_3314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3514_cast_fp16, var_3321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3518_cast_fp16, var_3328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3518_cast_fp16, var_3335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3518_cast_fp16, var_3342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3518_cast_fp16, var_3349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3522_cast_fp16, var_3356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3522_cast_fp16, var_3363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3522_cast_fp16, var_3370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3522_cast_fp16, var_3377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3526_cast_fp16, var_3384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3526_cast_fp16, var_3391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3526_cast_fp16, var_3398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3526_cast_fp16, var_3405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3530_cast_fp16, var_3412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3530_cast_fp16, var_3419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3530_cast_fp16, var_3426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3530_cast_fp16, var_3433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3534_cast_fp16, var_3440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3534_cast_fp16, var_3447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3534_cast_fp16, var_3454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3534_cast_fp16, var_3461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3538_cast_fp16, var_3468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3538_cast_fp16, var_3475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3538_cast_fp16, var_3482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; string _SplitHeadsQ__mh_w_383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_383_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_383_equation_0, values = (var_3538_cast_fp16, var_3489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_383_cast_fp16")]; fp16 var_3683_to_fp16 = const()[name = string("op_3683_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3683_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; fp16 var_3685_to_fp16 = const()[name = string("op_3685_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3685_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; fp16 var_3687_to_fp16 = const()[name = string("op_3687_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3687_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; fp16 var_3689_to_fp16 = const()[name = string("op_3689_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3689_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; fp16 var_3691_to_fp16 = const()[name = string("op_3691_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3691_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; fp16 var_3693_to_fp16 = const()[name = string("op_3693_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3693_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; fp16 var_3695_to_fp16 = const()[name = string("op_3695_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3695_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; fp16 var_3697_to_fp16 = const()[name = string("op_3697_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3697_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; fp16 var_3699_to_fp16 = const()[name = string("op_3699_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3699_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; fp16 var_3701_to_fp16 = const()[name = string("op_3701_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3701_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; fp16 var_3703_to_fp16 = const()[name = string("op_3703_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3703_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; fp16 var_3705_to_fp16 = const()[name = string("op_3705_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3705_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; fp16 var_3707_to_fp16 = const()[name = string("op_3707_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3707_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; fp16 var_3709_to_fp16 = const()[name = string("op_3709_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3709_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; fp16 var_3711_to_fp16 = const()[name = string("op_3711_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3711_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; fp16 var_3713_to_fp16 = const()[name = string("op_3713_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3713_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; fp16 var_3715_to_fp16 = const()[name = string("op_3715_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3715_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; fp16 var_3717_to_fp16 = const()[name = string("op_3717_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3717_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; fp16 var_3719_to_fp16 = const()[name = string("op_3719_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3719_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; fp16 var_3721_to_fp16 = const()[name = string("op_3721_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3721_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; fp16 var_3723_to_fp16 = const()[name = string("op_3723_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3723_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; fp16 var_3725_to_fp16 = const()[name = string("op_3725_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3725_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; fp16 var_3727_to_fp16 = const()[name = string("op_3727_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3727_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; fp16 var_3729_to_fp16 = const()[name = string("op_3729_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3729_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; fp16 var_3731_to_fp16 = const()[name = string("op_3731_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3731_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; fp16 var_3733_to_fp16 = const()[name = string("op_3733_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3733_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; fp16 var_3735_to_fp16 = const()[name = string("op_3735_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3735_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; fp16 var_3737_to_fp16 = const()[name = string("op_3737_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3737_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; fp16 var_3739_to_fp16 = const()[name = string("op_3739_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3739_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; fp16 var_3741_to_fp16 = const()[name = string("op_3741_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3741_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; fp16 var_3743_to_fp16 = const()[name = string("op_3743_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3743_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; fp16 var_3745_to_fp16 = const()[name = string("op_3745_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3745_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; fp16 var_3747_to_fp16 = const()[name = string("op_3747_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3747_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; fp16 var_3749_to_fp16 = const()[name = string("op_3749_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3749_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; fp16 var_3751_to_fp16 = const()[name = string("op_3751_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3751_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; fp16 var_3753_to_fp16 = const()[name = string("op_3753_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3753_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; fp16 var_3755_to_fp16 = const()[name = string("op_3755_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_3755_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; fp16 var_3757_to_fp16 = const()[name = string("op_3757_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_3757_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; fp16 var_3759_to_fp16 = const()[name = string("op_3759_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_3759_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; fp16 var_3761_to_fp16 = const()[name = string("op_3761_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_3761_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; fp16 var_3763_to_fp16 = const()[name = string("op_3763_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_3763_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; fp16 var_3765_to_fp16 = const()[name = string("op_3765_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_3765_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; fp16 var_3767_to_fp16 = const()[name = string("op_3767_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_3767_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; fp16 var_3769_to_fp16 = const()[name = string("op_3769_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_3769_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; fp16 var_3771_to_fp16 = const()[name = string("op_3771_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_3771_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; fp16 var_3773_to_fp16 = const()[name = string("op_3773_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_3773_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; fp16 var_3775_to_fp16 = const()[name = string("op_3775_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_3775_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; fp16 var_3777_to_fp16 = const()[name = string("op_3777_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_383_cast_fp16, y = var_3777_to_fp16)[name = string("aw_chunk_383_cast_fp16")]; tensor var_3779_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_289_cast_fp16)[name = string("op_3779_cast_fp16")]; tensor var_3780_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_291_cast_fp16)[name = string("op_3780_cast_fp16")]; tensor var_3781_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_293_cast_fp16)[name = string("op_3781_cast_fp16")]; tensor var_3782_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_295_cast_fp16)[name = string("op_3782_cast_fp16")]; tensor var_3783_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_297_cast_fp16)[name = string("op_3783_cast_fp16")]; tensor var_3784_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_299_cast_fp16)[name = string("op_3784_cast_fp16")]; tensor var_3785_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_301_cast_fp16)[name = string("op_3785_cast_fp16")]; tensor var_3786_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_303_cast_fp16)[name = string("op_3786_cast_fp16")]; tensor var_3787_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_305_cast_fp16)[name = string("op_3787_cast_fp16")]; tensor var_3788_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_307_cast_fp16)[name = string("op_3788_cast_fp16")]; tensor var_3789_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_309_cast_fp16)[name = string("op_3789_cast_fp16")]; tensor var_3790_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_311_cast_fp16)[name = string("op_3790_cast_fp16")]; tensor var_3791_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_313_cast_fp16)[name = string("op_3791_cast_fp16")]; tensor var_3792_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_315_cast_fp16)[name = string("op_3792_cast_fp16")]; tensor var_3793_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_317_cast_fp16)[name = string("op_3793_cast_fp16")]; tensor var_3794_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_319_cast_fp16)[name = string("op_3794_cast_fp16")]; tensor var_3795_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_321_cast_fp16)[name = string("op_3795_cast_fp16")]; tensor var_3796_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_323_cast_fp16)[name = string("op_3796_cast_fp16")]; tensor var_3797_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_325_cast_fp16)[name = string("op_3797_cast_fp16")]; tensor var_3798_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_327_cast_fp16)[name = string("op_3798_cast_fp16")]; tensor var_3799_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_329_cast_fp16)[name = string("op_3799_cast_fp16")]; tensor var_3800_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_331_cast_fp16)[name = string("op_3800_cast_fp16")]; tensor var_3801_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_333_cast_fp16)[name = string("op_3801_cast_fp16")]; tensor var_3802_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_335_cast_fp16)[name = string("op_3802_cast_fp16")]; tensor var_3803_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_337_cast_fp16)[name = string("op_3803_cast_fp16")]; tensor var_3804_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_339_cast_fp16)[name = string("op_3804_cast_fp16")]; tensor var_3805_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_341_cast_fp16)[name = string("op_3805_cast_fp16")]; tensor var_3806_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_343_cast_fp16)[name = string("op_3806_cast_fp16")]; tensor var_3807_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_345_cast_fp16)[name = string("op_3807_cast_fp16")]; tensor var_3808_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_347_cast_fp16)[name = string("op_3808_cast_fp16")]; tensor var_3809_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_349_cast_fp16)[name = string("op_3809_cast_fp16")]; tensor var_3810_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_351_cast_fp16)[name = string("op_3810_cast_fp16")]; tensor var_3811_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_353_cast_fp16)[name = string("op_3811_cast_fp16")]; tensor var_3812_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_355_cast_fp16)[name = string("op_3812_cast_fp16")]; tensor var_3813_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_357_cast_fp16)[name = string("op_3813_cast_fp16")]; tensor var_3814_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_359_cast_fp16)[name = string("op_3814_cast_fp16")]; tensor var_3815_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_361_cast_fp16)[name = string("op_3815_cast_fp16")]; tensor var_3816_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_363_cast_fp16)[name = string("op_3816_cast_fp16")]; tensor var_3817_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_365_cast_fp16)[name = string("op_3817_cast_fp16")]; tensor var_3818_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_367_cast_fp16)[name = string("op_3818_cast_fp16")]; tensor var_3819_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_369_cast_fp16)[name = string("op_3819_cast_fp16")]; tensor var_3820_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_371_cast_fp16)[name = string("op_3820_cast_fp16")]; tensor var_3821_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_373_cast_fp16)[name = string("op_3821_cast_fp16")]; tensor var_3822_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_375_cast_fp16)[name = string("op_3822_cast_fp16")]; tensor var_3823_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_377_cast_fp16)[name = string("op_3823_cast_fp16")]; tensor var_3824_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_379_cast_fp16)[name = string("op_3824_cast_fp16")]; tensor var_3825_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_381_cast_fp16)[name = string("op_3825_cast_fp16")]; tensor var_3826_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_383_cast_fp16)[name = string("op_3826_cast_fp16")]; string var_3828_equation_0 = const()[name = string("op_3828_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3828_cast_fp16 = einsum(equation = var_3828_equation_0, values = (var_3540_cast_fp16, var_3779_cast_fp16))[name = string("op_3828_cast_fp16")]; string var_3830_equation_0 = const()[name = string("op_3830_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3830_cast_fp16 = einsum(equation = var_3830_equation_0, values = (var_3540_cast_fp16, var_3780_cast_fp16))[name = string("op_3830_cast_fp16")]; string var_3832_equation_0 = const()[name = string("op_3832_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3832_cast_fp16 = einsum(equation = var_3832_equation_0, values = (var_3540_cast_fp16, var_3781_cast_fp16))[name = string("op_3832_cast_fp16")]; string var_3834_equation_0 = const()[name = string("op_3834_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3834_cast_fp16 = einsum(equation = var_3834_equation_0, values = (var_3540_cast_fp16, var_3782_cast_fp16))[name = string("op_3834_cast_fp16")]; string var_3836_equation_0 = const()[name = string("op_3836_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3836_cast_fp16 = einsum(equation = var_3836_equation_0, values = (var_3544_cast_fp16, var_3783_cast_fp16))[name = string("op_3836_cast_fp16")]; string var_3838_equation_0 = const()[name = string("op_3838_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3838_cast_fp16 = einsum(equation = var_3838_equation_0, values = (var_3544_cast_fp16, var_3784_cast_fp16))[name = string("op_3838_cast_fp16")]; string var_3840_equation_0 = const()[name = string("op_3840_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3840_cast_fp16 = einsum(equation = var_3840_equation_0, values = (var_3544_cast_fp16, var_3785_cast_fp16))[name = string("op_3840_cast_fp16")]; string var_3842_equation_0 = const()[name = string("op_3842_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3842_cast_fp16 = einsum(equation = var_3842_equation_0, values = (var_3544_cast_fp16, var_3786_cast_fp16))[name = string("op_3842_cast_fp16")]; string var_3844_equation_0 = const()[name = string("op_3844_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3844_cast_fp16 = einsum(equation = var_3844_equation_0, values = (var_3548_cast_fp16, var_3787_cast_fp16))[name = string("op_3844_cast_fp16")]; string var_3846_equation_0 = const()[name = string("op_3846_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3846_cast_fp16 = einsum(equation = var_3846_equation_0, values = (var_3548_cast_fp16, var_3788_cast_fp16))[name = string("op_3846_cast_fp16")]; string var_3848_equation_0 = const()[name = string("op_3848_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3848_cast_fp16 = einsum(equation = var_3848_equation_0, values = (var_3548_cast_fp16, var_3789_cast_fp16))[name = string("op_3848_cast_fp16")]; string var_3850_equation_0 = const()[name = string("op_3850_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3850_cast_fp16 = einsum(equation = var_3850_equation_0, values = (var_3548_cast_fp16, var_3790_cast_fp16))[name = string("op_3850_cast_fp16")]; string var_3852_equation_0 = const()[name = string("op_3852_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3852_cast_fp16 = einsum(equation = var_3852_equation_0, values = (var_3552_cast_fp16, var_3791_cast_fp16))[name = string("op_3852_cast_fp16")]; string var_3854_equation_0 = const()[name = string("op_3854_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3854_cast_fp16 = einsum(equation = var_3854_equation_0, values = (var_3552_cast_fp16, var_3792_cast_fp16))[name = string("op_3854_cast_fp16")]; string var_3856_equation_0 = const()[name = string("op_3856_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3856_cast_fp16 = einsum(equation = var_3856_equation_0, values = (var_3552_cast_fp16, var_3793_cast_fp16))[name = string("op_3856_cast_fp16")]; string var_3858_equation_0 = const()[name = string("op_3858_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3858_cast_fp16 = einsum(equation = var_3858_equation_0, values = (var_3552_cast_fp16, var_3794_cast_fp16))[name = string("op_3858_cast_fp16")]; string var_3860_equation_0 = const()[name = string("op_3860_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3860_cast_fp16 = einsum(equation = var_3860_equation_0, values = (var_3556_cast_fp16, var_3795_cast_fp16))[name = string("op_3860_cast_fp16")]; string var_3862_equation_0 = const()[name = string("op_3862_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3862_cast_fp16 = einsum(equation = var_3862_equation_0, values = (var_3556_cast_fp16, var_3796_cast_fp16))[name = string("op_3862_cast_fp16")]; string var_3864_equation_0 = const()[name = string("op_3864_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3864_cast_fp16 = einsum(equation = var_3864_equation_0, values = (var_3556_cast_fp16, var_3797_cast_fp16))[name = string("op_3864_cast_fp16")]; string var_3866_equation_0 = const()[name = string("op_3866_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3866_cast_fp16 = einsum(equation = var_3866_equation_0, values = (var_3556_cast_fp16, var_3798_cast_fp16))[name = string("op_3866_cast_fp16")]; string var_3868_equation_0 = const()[name = string("op_3868_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3868_cast_fp16 = einsum(equation = var_3868_equation_0, values = (var_3560_cast_fp16, var_3799_cast_fp16))[name = string("op_3868_cast_fp16")]; string var_3870_equation_0 = const()[name = string("op_3870_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3870_cast_fp16 = einsum(equation = var_3870_equation_0, values = (var_3560_cast_fp16, var_3800_cast_fp16))[name = string("op_3870_cast_fp16")]; string var_3872_equation_0 = const()[name = string("op_3872_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3872_cast_fp16 = einsum(equation = var_3872_equation_0, values = (var_3560_cast_fp16, var_3801_cast_fp16))[name = string("op_3872_cast_fp16")]; string var_3874_equation_0 = const()[name = string("op_3874_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3874_cast_fp16 = einsum(equation = var_3874_equation_0, values = (var_3560_cast_fp16, var_3802_cast_fp16))[name = string("op_3874_cast_fp16")]; string var_3876_equation_0 = const()[name = string("op_3876_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3876_cast_fp16 = einsum(equation = var_3876_equation_0, values = (var_3564_cast_fp16, var_3803_cast_fp16))[name = string("op_3876_cast_fp16")]; string var_3878_equation_0 = const()[name = string("op_3878_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3878_cast_fp16 = einsum(equation = var_3878_equation_0, values = (var_3564_cast_fp16, var_3804_cast_fp16))[name = string("op_3878_cast_fp16")]; string var_3880_equation_0 = const()[name = string("op_3880_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3880_cast_fp16 = einsum(equation = var_3880_equation_0, values = (var_3564_cast_fp16, var_3805_cast_fp16))[name = string("op_3880_cast_fp16")]; string var_3882_equation_0 = const()[name = string("op_3882_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3882_cast_fp16 = einsum(equation = var_3882_equation_0, values = (var_3564_cast_fp16, var_3806_cast_fp16))[name = string("op_3882_cast_fp16")]; string var_3884_equation_0 = const()[name = string("op_3884_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3884_cast_fp16 = einsum(equation = var_3884_equation_0, values = (var_3568_cast_fp16, var_3807_cast_fp16))[name = string("op_3884_cast_fp16")]; string var_3886_equation_0 = const()[name = string("op_3886_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3886_cast_fp16 = einsum(equation = var_3886_equation_0, values = (var_3568_cast_fp16, var_3808_cast_fp16))[name = string("op_3886_cast_fp16")]; string var_3888_equation_0 = const()[name = string("op_3888_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3888_cast_fp16 = einsum(equation = var_3888_equation_0, values = (var_3568_cast_fp16, var_3809_cast_fp16))[name = string("op_3888_cast_fp16")]; string var_3890_equation_0 = const()[name = string("op_3890_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3890_cast_fp16 = einsum(equation = var_3890_equation_0, values = (var_3568_cast_fp16, var_3810_cast_fp16))[name = string("op_3890_cast_fp16")]; string var_3892_equation_0 = const()[name = string("op_3892_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3892_cast_fp16 = einsum(equation = var_3892_equation_0, values = (var_3572_cast_fp16, var_3811_cast_fp16))[name = string("op_3892_cast_fp16")]; string var_3894_equation_0 = const()[name = string("op_3894_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3894_cast_fp16 = einsum(equation = var_3894_equation_0, values = (var_3572_cast_fp16, var_3812_cast_fp16))[name = string("op_3894_cast_fp16")]; string var_3896_equation_0 = const()[name = string("op_3896_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3896_cast_fp16 = einsum(equation = var_3896_equation_0, values = (var_3572_cast_fp16, var_3813_cast_fp16))[name = string("op_3896_cast_fp16")]; string var_3898_equation_0 = const()[name = string("op_3898_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3898_cast_fp16 = einsum(equation = var_3898_equation_0, values = (var_3572_cast_fp16, var_3814_cast_fp16))[name = string("op_3898_cast_fp16")]; string var_3900_equation_0 = const()[name = string("op_3900_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3900_cast_fp16 = einsum(equation = var_3900_equation_0, values = (var_3576_cast_fp16, var_3815_cast_fp16))[name = string("op_3900_cast_fp16")]; string var_3902_equation_0 = const()[name = string("op_3902_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3902_cast_fp16 = einsum(equation = var_3902_equation_0, values = (var_3576_cast_fp16, var_3816_cast_fp16))[name = string("op_3902_cast_fp16")]; string var_3904_equation_0 = const()[name = string("op_3904_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3904_cast_fp16 = einsum(equation = var_3904_equation_0, values = (var_3576_cast_fp16, var_3817_cast_fp16))[name = string("op_3904_cast_fp16")]; string var_3906_equation_0 = const()[name = string("op_3906_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3906_cast_fp16 = einsum(equation = var_3906_equation_0, values = (var_3576_cast_fp16, var_3818_cast_fp16))[name = string("op_3906_cast_fp16")]; string var_3908_equation_0 = const()[name = string("op_3908_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3908_cast_fp16 = einsum(equation = var_3908_equation_0, values = (var_3580_cast_fp16, var_3819_cast_fp16))[name = string("op_3908_cast_fp16")]; string var_3910_equation_0 = const()[name = string("op_3910_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3910_cast_fp16 = einsum(equation = var_3910_equation_0, values = (var_3580_cast_fp16, var_3820_cast_fp16))[name = string("op_3910_cast_fp16")]; string var_3912_equation_0 = const()[name = string("op_3912_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3912_cast_fp16 = einsum(equation = var_3912_equation_0, values = (var_3580_cast_fp16, var_3821_cast_fp16))[name = string("op_3912_cast_fp16")]; string var_3914_equation_0 = const()[name = string("op_3914_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3914_cast_fp16 = einsum(equation = var_3914_equation_0, values = (var_3580_cast_fp16, var_3822_cast_fp16))[name = string("op_3914_cast_fp16")]; string var_3916_equation_0 = const()[name = string("op_3916_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3916_cast_fp16 = einsum(equation = var_3916_equation_0, values = (var_3584_cast_fp16, var_3823_cast_fp16))[name = string("op_3916_cast_fp16")]; string var_3918_equation_0 = const()[name = string("op_3918_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3918_cast_fp16 = einsum(equation = var_3918_equation_0, values = (var_3584_cast_fp16, var_3824_cast_fp16))[name = string("op_3918_cast_fp16")]; string var_3920_equation_0 = const()[name = string("op_3920_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3920_cast_fp16 = einsum(equation = var_3920_equation_0, values = (var_3584_cast_fp16, var_3825_cast_fp16))[name = string("op_3920_cast_fp16")]; string var_3922_equation_0 = const()[name = string("op_3922_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_3922_cast_fp16 = einsum(equation = var_3922_equation_0, values = (var_3584_cast_fp16, var_3826_cast_fp16))[name = string("op_3922_cast_fp16")]; bool var_3924_interleave_0 = const()[name = string("op_3924_interleave_0"), val = bool(false)]; tensor var_3924_cast_fp16 = concat(axis = var_3035, interleave = var_3924_interleave_0, values = (var_3828_cast_fp16, var_3830_cast_fp16, var_3832_cast_fp16, var_3834_cast_fp16))[name = string("op_3924_cast_fp16")]; bool var_3926_interleave_0 = const()[name = string("op_3926_interleave_0"), val = bool(false)]; tensor var_3926_cast_fp16 = concat(axis = var_3035, interleave = var_3926_interleave_0, values = (var_3836_cast_fp16, var_3838_cast_fp16, var_3840_cast_fp16, var_3842_cast_fp16))[name = string("op_3926_cast_fp16")]; bool var_3928_interleave_0 = const()[name = string("op_3928_interleave_0"), val = bool(false)]; tensor var_3928_cast_fp16 = concat(axis = var_3035, interleave = var_3928_interleave_0, values = (var_3844_cast_fp16, var_3846_cast_fp16, var_3848_cast_fp16, var_3850_cast_fp16))[name = string("op_3928_cast_fp16")]; bool var_3930_interleave_0 = const()[name = string("op_3930_interleave_0"), val = bool(false)]; tensor var_3930_cast_fp16 = concat(axis = var_3035, interleave = var_3930_interleave_0, values = (var_3852_cast_fp16, var_3854_cast_fp16, var_3856_cast_fp16, var_3858_cast_fp16))[name = string("op_3930_cast_fp16")]; bool var_3932_interleave_0 = const()[name = string("op_3932_interleave_0"), val = bool(false)]; tensor var_3932_cast_fp16 = concat(axis = var_3035, interleave = var_3932_interleave_0, values = (var_3860_cast_fp16, var_3862_cast_fp16, var_3864_cast_fp16, var_3866_cast_fp16))[name = string("op_3932_cast_fp16")]; bool var_3934_interleave_0 = const()[name = string("op_3934_interleave_0"), val = bool(false)]; tensor var_3934_cast_fp16 = concat(axis = var_3035, interleave = var_3934_interleave_0, values = (var_3868_cast_fp16, var_3870_cast_fp16, var_3872_cast_fp16, var_3874_cast_fp16))[name = string("op_3934_cast_fp16")]; bool var_3936_interleave_0 = const()[name = string("op_3936_interleave_0"), val = bool(false)]; tensor var_3936_cast_fp16 = concat(axis = var_3035, interleave = var_3936_interleave_0, values = (var_3876_cast_fp16, var_3878_cast_fp16, var_3880_cast_fp16, var_3882_cast_fp16))[name = string("op_3936_cast_fp16")]; bool var_3938_interleave_0 = const()[name = string("op_3938_interleave_0"), val = bool(false)]; tensor var_3938_cast_fp16 = concat(axis = var_3035, interleave = var_3938_interleave_0, values = (var_3884_cast_fp16, var_3886_cast_fp16, var_3888_cast_fp16, var_3890_cast_fp16))[name = string("op_3938_cast_fp16")]; bool var_3940_interleave_0 = const()[name = string("op_3940_interleave_0"), val = bool(false)]; tensor var_3940_cast_fp16 = concat(axis = var_3035, interleave = var_3940_interleave_0, values = (var_3892_cast_fp16, var_3894_cast_fp16, var_3896_cast_fp16, var_3898_cast_fp16))[name = string("op_3940_cast_fp16")]; bool var_3942_interleave_0 = const()[name = string("op_3942_interleave_0"), val = bool(false)]; tensor var_3942_cast_fp16 = concat(axis = var_3035, interleave = var_3942_interleave_0, values = (var_3900_cast_fp16, var_3902_cast_fp16, var_3904_cast_fp16, var_3906_cast_fp16))[name = string("op_3942_cast_fp16")]; bool var_3944_interleave_0 = const()[name = string("op_3944_interleave_0"), val = bool(false)]; tensor var_3944_cast_fp16 = concat(axis = var_3035, interleave = var_3944_interleave_0, values = (var_3908_cast_fp16, var_3910_cast_fp16, var_3912_cast_fp16, var_3914_cast_fp16))[name = string("op_3944_cast_fp16")]; bool var_3946_interleave_0 = const()[name = string("op_3946_interleave_0"), val = bool(false)]; tensor var_3946_cast_fp16 = concat(axis = var_3035, interleave = var_3946_interleave_0, values = (var_3916_cast_fp16, var_3918_cast_fp16, var_3920_cast_fp16, var_3922_cast_fp16))[name = string("op_3946_cast_fp16")]; bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; tensor input_25_cast_fp16 = concat(axis = var_3052, interleave = input_25_interleave_0, values = (var_3924_cast_fp16, var_3926_cast_fp16, var_3928_cast_fp16, var_3930_cast_fp16, var_3932_cast_fp16, var_3934_cast_fp16, var_3936_cast_fp16, var_3938_cast_fp16, var_3940_cast_fp16, var_3942_cast_fp16, var_3944_cast_fp16, var_3946_cast_fp16))[name = string("input_25_cast_fp16")]; string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52289280)))]; tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53468992)))]; tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_3965_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53470592)))]; tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53472192)))]; fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53473792)))]; tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58192448)))]; tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58198656)))]; tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62917312)))]; tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; int32 var_3994 = const()[name = string("op_3994"), val = int32(3)]; int32 var_4011 = const()[name = string("op_4011"), val = int32(1)]; tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; fp16 var_4028_to_fp16 = const()[name = string("op_4028_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_4028_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62918912)))]; tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62920512)))]; fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62922112)))]; tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64101824)))]; tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64103424)))]; tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65283136)))]; tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66462848)))]; tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; tensor var_4066_begin_0 = const()[name = string("op_4066_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4066_end_0 = const()[name = string("op_4066_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4066_end_mask_0 = const()[name = string("op_4066_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4066_cast_fp16 = slice_by_index(begin = var_4066_begin_0, end = var_4066_end_0, end_mask = var_4066_end_mask_0, x = query_9_cast_fp16)[name = string("op_4066_cast_fp16")]; tensor var_4070_begin_0 = const()[name = string("op_4070_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_4070_end_0 = const()[name = string("op_4070_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_4070_end_mask_0 = const()[name = string("op_4070_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4070_cast_fp16 = slice_by_index(begin = var_4070_begin_0, end = var_4070_end_0, end_mask = var_4070_end_mask_0, x = query_9_cast_fp16)[name = string("op_4070_cast_fp16")]; tensor var_4074_begin_0 = const()[name = string("op_4074_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_4074_end_0 = const()[name = string("op_4074_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_4074_end_mask_0 = const()[name = string("op_4074_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4074_cast_fp16 = slice_by_index(begin = var_4074_begin_0, end = var_4074_end_0, end_mask = var_4074_end_mask_0, x = query_9_cast_fp16)[name = string("op_4074_cast_fp16")]; tensor var_4078_begin_0 = const()[name = string("op_4078_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_4078_end_0 = const()[name = string("op_4078_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_4078_end_mask_0 = const()[name = string("op_4078_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4078_cast_fp16 = slice_by_index(begin = var_4078_begin_0, end = var_4078_end_0, end_mask = var_4078_end_mask_0, x = query_9_cast_fp16)[name = string("op_4078_cast_fp16")]; tensor var_4082_begin_0 = const()[name = string("op_4082_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_4082_end_0 = const()[name = string("op_4082_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_4082_end_mask_0 = const()[name = string("op_4082_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4082_cast_fp16 = slice_by_index(begin = var_4082_begin_0, end = var_4082_end_0, end_mask = var_4082_end_mask_0, x = query_9_cast_fp16)[name = string("op_4082_cast_fp16")]; tensor var_4086_begin_0 = const()[name = string("op_4086_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4086_end_0 = const()[name = string("op_4086_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_4086_end_mask_0 = const()[name = string("op_4086_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4086_cast_fp16 = slice_by_index(begin = var_4086_begin_0, end = var_4086_end_0, end_mask = var_4086_end_mask_0, x = query_9_cast_fp16)[name = string("op_4086_cast_fp16")]; tensor var_4090_begin_0 = const()[name = string("op_4090_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_4090_end_0 = const()[name = string("op_4090_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_4090_end_mask_0 = const()[name = string("op_4090_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4090_cast_fp16 = slice_by_index(begin = var_4090_begin_0, end = var_4090_end_0, end_mask = var_4090_end_mask_0, x = query_9_cast_fp16)[name = string("op_4090_cast_fp16")]; tensor var_4094_begin_0 = const()[name = string("op_4094_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_4094_end_0 = const()[name = string("op_4094_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_4094_end_mask_0 = const()[name = string("op_4094_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4094_cast_fp16 = slice_by_index(begin = var_4094_begin_0, end = var_4094_end_0, end_mask = var_4094_end_mask_0, x = query_9_cast_fp16)[name = string("op_4094_cast_fp16")]; tensor var_4098_begin_0 = const()[name = string("op_4098_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_4098_end_0 = const()[name = string("op_4098_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_4098_end_mask_0 = const()[name = string("op_4098_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4098_cast_fp16 = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, x = query_9_cast_fp16)[name = string("op_4098_cast_fp16")]; tensor var_4102_begin_0 = const()[name = string("op_4102_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_4102_end_0 = const()[name = string("op_4102_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_4102_end_mask_0 = const()[name = string("op_4102_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4102_cast_fp16 = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = query_9_cast_fp16)[name = string("op_4102_cast_fp16")]; tensor var_4106_begin_0 = const()[name = string("op_4106_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_4106_end_0 = const()[name = string("op_4106_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_4106_end_mask_0 = const()[name = string("op_4106_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4106_cast_fp16 = slice_by_index(begin = var_4106_begin_0, end = var_4106_end_0, end_mask = var_4106_end_mask_0, x = query_9_cast_fp16)[name = string("op_4106_cast_fp16")]; tensor var_4110_begin_0 = const()[name = string("op_4110_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_4110_end_0 = const()[name = string("op_4110_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_4110_end_mask_0 = const()[name = string("op_4110_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4110_cast_fp16 = slice_by_index(begin = var_4110_begin_0, end = var_4110_end_0, end_mask = var_4110_end_mask_0, x = query_9_cast_fp16)[name = string("op_4110_cast_fp16")]; tensor var_4119_begin_0 = const()[name = string("op_4119_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4119_end_0 = const()[name = string("op_4119_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4119_end_mask_0 = const()[name = string("op_4119_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4119_cast_fp16 = slice_by_index(begin = var_4119_begin_0, end = var_4119_end_0, end_mask = var_4119_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4119_cast_fp16")]; tensor var_4126_begin_0 = const()[name = string("op_4126_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4126_end_0 = const()[name = string("op_4126_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4126_end_mask_0 = const()[name = string("op_4126_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4126_cast_fp16 = slice_by_index(begin = var_4126_begin_0, end = var_4126_end_0, end_mask = var_4126_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4126_cast_fp16")]; tensor var_4133_begin_0 = const()[name = string("op_4133_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4133_end_0 = const()[name = string("op_4133_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4133_end_mask_0 = const()[name = string("op_4133_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4133_cast_fp16 = slice_by_index(begin = var_4133_begin_0, end = var_4133_end_0, end_mask = var_4133_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4133_cast_fp16")]; tensor var_4140_begin_0 = const()[name = string("op_4140_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4140_end_0 = const()[name = string("op_4140_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4140_end_mask_0 = const()[name = string("op_4140_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4140_cast_fp16 = slice_by_index(begin = var_4140_begin_0, end = var_4140_end_0, end_mask = var_4140_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4140_cast_fp16")]; tensor var_4147_begin_0 = const()[name = string("op_4147_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4147_end_0 = const()[name = string("op_4147_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4147_end_mask_0 = const()[name = string("op_4147_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4147_cast_fp16 = slice_by_index(begin = var_4147_begin_0, end = var_4147_end_0, end_mask = var_4147_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4147_cast_fp16")]; tensor var_4154_begin_0 = const()[name = string("op_4154_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4154_end_0 = const()[name = string("op_4154_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4154_end_mask_0 = const()[name = string("op_4154_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4154_cast_fp16 = slice_by_index(begin = var_4154_begin_0, end = var_4154_end_0, end_mask = var_4154_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4154_cast_fp16")]; tensor var_4161_begin_0 = const()[name = string("op_4161_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4161_end_0 = const()[name = string("op_4161_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4161_end_mask_0 = const()[name = string("op_4161_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4161_cast_fp16 = slice_by_index(begin = var_4161_begin_0, end = var_4161_end_0, end_mask = var_4161_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4161_cast_fp16")]; tensor var_4168_begin_0 = const()[name = string("op_4168_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4168_end_0 = const()[name = string("op_4168_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4168_end_mask_0 = const()[name = string("op_4168_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4168_cast_fp16 = slice_by_index(begin = var_4168_begin_0, end = var_4168_end_0, end_mask = var_4168_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4168_cast_fp16")]; tensor var_4175_begin_0 = const()[name = string("op_4175_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4175_end_0 = const()[name = string("op_4175_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4175_end_mask_0 = const()[name = string("op_4175_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4175_cast_fp16 = slice_by_index(begin = var_4175_begin_0, end = var_4175_end_0, end_mask = var_4175_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4175_cast_fp16")]; tensor var_4182_begin_0 = const()[name = string("op_4182_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4182_end_0 = const()[name = string("op_4182_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4182_end_mask_0 = const()[name = string("op_4182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4182_cast_fp16 = slice_by_index(begin = var_4182_begin_0, end = var_4182_end_0, end_mask = var_4182_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4182_cast_fp16")]; tensor var_4189_begin_0 = const()[name = string("op_4189_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4189_end_0 = const()[name = string("op_4189_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4189_end_mask_0 = const()[name = string("op_4189_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4189_cast_fp16 = slice_by_index(begin = var_4189_begin_0, end = var_4189_end_0, end_mask = var_4189_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4189_cast_fp16")]; tensor var_4196_begin_0 = const()[name = string("op_4196_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4196_end_0 = const()[name = string("op_4196_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4196_end_mask_0 = const()[name = string("op_4196_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4196_cast_fp16 = slice_by_index(begin = var_4196_begin_0, end = var_4196_end_0, end_mask = var_4196_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4196_cast_fp16")]; tensor var_4203_begin_0 = const()[name = string("op_4203_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4203_end_0 = const()[name = string("op_4203_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4203_end_mask_0 = const()[name = string("op_4203_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4203_cast_fp16 = slice_by_index(begin = var_4203_begin_0, end = var_4203_end_0, end_mask = var_4203_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4203_cast_fp16")]; tensor var_4210_begin_0 = const()[name = string("op_4210_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4210_end_0 = const()[name = string("op_4210_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4210_end_mask_0 = const()[name = string("op_4210_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4210_cast_fp16 = slice_by_index(begin = var_4210_begin_0, end = var_4210_end_0, end_mask = var_4210_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4210_cast_fp16")]; tensor var_4217_begin_0 = const()[name = string("op_4217_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4217_end_0 = const()[name = string("op_4217_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4217_end_mask_0 = const()[name = string("op_4217_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4217_cast_fp16 = slice_by_index(begin = var_4217_begin_0, end = var_4217_end_0, end_mask = var_4217_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4217_cast_fp16")]; tensor var_4224_begin_0 = const()[name = string("op_4224_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4224_end_0 = const()[name = string("op_4224_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4224_end_mask_0 = const()[name = string("op_4224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4224_cast_fp16 = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4224_cast_fp16")]; tensor var_4231_begin_0 = const()[name = string("op_4231_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4231_end_0 = const()[name = string("op_4231_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4231_end_mask_0 = const()[name = string("op_4231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4231_cast_fp16 = slice_by_index(begin = var_4231_begin_0, end = var_4231_end_0, end_mask = var_4231_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4231_cast_fp16")]; tensor var_4238_begin_0 = const()[name = string("op_4238_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4238_end_0 = const()[name = string("op_4238_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4238_end_mask_0 = const()[name = string("op_4238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4238_cast_fp16")]; tensor var_4245_begin_0 = const()[name = string("op_4245_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4245_end_0 = const()[name = string("op_4245_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4245_end_mask_0 = const()[name = string("op_4245_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4245_cast_fp16 = slice_by_index(begin = var_4245_begin_0, end = var_4245_end_0, end_mask = var_4245_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4245_cast_fp16")]; tensor var_4252_begin_0 = const()[name = string("op_4252_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4252_end_0 = const()[name = string("op_4252_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4252_end_mask_0 = const()[name = string("op_4252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4252_cast_fp16 = slice_by_index(begin = var_4252_begin_0, end = var_4252_end_0, end_mask = var_4252_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4252_cast_fp16")]; tensor var_4259_begin_0 = const()[name = string("op_4259_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4259_end_0 = const()[name = string("op_4259_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4259_end_mask_0 = const()[name = string("op_4259_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4259_cast_fp16 = slice_by_index(begin = var_4259_begin_0, end = var_4259_end_0, end_mask = var_4259_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4259_cast_fp16")]; tensor var_4266_begin_0 = const()[name = string("op_4266_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4266_end_0 = const()[name = string("op_4266_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4266_end_mask_0 = const()[name = string("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4266_cast_fp16")]; tensor var_4273_begin_0 = const()[name = string("op_4273_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4273_end_0 = const()[name = string("op_4273_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4273_end_mask_0 = const()[name = string("op_4273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4273_cast_fp16")]; tensor var_4280_begin_0 = const()[name = string("op_4280_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4280_end_0 = const()[name = string("op_4280_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4280_end_mask_0 = const()[name = string("op_4280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4280_cast_fp16")]; tensor var_4287_begin_0 = const()[name = string("op_4287_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4287_end_0 = const()[name = string("op_4287_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4287_end_mask_0 = const()[name = string("op_4287_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4287_cast_fp16 = slice_by_index(begin = var_4287_begin_0, end = var_4287_end_0, end_mask = var_4287_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4287_cast_fp16")]; tensor var_4294_begin_0 = const()[name = string("op_4294_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4294_end_0 = const()[name = string("op_4294_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4294_end_mask_0 = const()[name = string("op_4294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4294_cast_fp16 = slice_by_index(begin = var_4294_begin_0, end = var_4294_end_0, end_mask = var_4294_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4294_cast_fp16")]; tensor var_4301_begin_0 = const()[name = string("op_4301_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4301_end_0 = const()[name = string("op_4301_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4301_end_mask_0 = const()[name = string("op_4301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4301_cast_fp16 = slice_by_index(begin = var_4301_begin_0, end = var_4301_end_0, end_mask = var_4301_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4301_cast_fp16")]; tensor var_4308_begin_0 = const()[name = string("op_4308_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4308_end_0 = const()[name = string("op_4308_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4308_end_mask_0 = const()[name = string("op_4308_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4308_cast_fp16 = slice_by_index(begin = var_4308_begin_0, end = var_4308_end_0, end_mask = var_4308_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4308_cast_fp16")]; tensor var_4315_begin_0 = const()[name = string("op_4315_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4315_end_0 = const()[name = string("op_4315_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4315_end_mask_0 = const()[name = string("op_4315_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4315_cast_fp16 = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4315_cast_fp16")]; tensor var_4322_begin_0 = const()[name = string("op_4322_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4322_end_0 = const()[name = string("op_4322_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4322_end_mask_0 = const()[name = string("op_4322_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4322_cast_fp16 = slice_by_index(begin = var_4322_begin_0, end = var_4322_end_0, end_mask = var_4322_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4322_cast_fp16")]; tensor var_4329_begin_0 = const()[name = string("op_4329_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4329_end_0 = const()[name = string("op_4329_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4329_end_mask_0 = const()[name = string("op_4329_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4329_cast_fp16 = slice_by_index(begin = var_4329_begin_0, end = var_4329_end_0, end_mask = var_4329_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4329_cast_fp16")]; tensor var_4336_begin_0 = const()[name = string("op_4336_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4336_end_0 = const()[name = string("op_4336_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4336_end_mask_0 = const()[name = string("op_4336_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4336_cast_fp16 = slice_by_index(begin = var_4336_begin_0, end = var_4336_end_0, end_mask = var_4336_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4336_cast_fp16")]; tensor var_4343_begin_0 = const()[name = string("op_4343_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4343_end_0 = const()[name = string("op_4343_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4343_end_mask_0 = const()[name = string("op_4343_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4343_cast_fp16")]; tensor var_4350_begin_0 = const()[name = string("op_4350_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4350_end_0 = const()[name = string("op_4350_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4350_end_mask_0 = const()[name = string("op_4350_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4350_cast_fp16 = slice_by_index(begin = var_4350_begin_0, end = var_4350_end_0, end_mask = var_4350_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4350_cast_fp16")]; tensor var_4357_begin_0 = const()[name = string("op_4357_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4357_end_0 = const()[name = string("op_4357_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4357_end_mask_0 = const()[name = string("op_4357_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4357_cast_fp16 = slice_by_index(begin = var_4357_begin_0, end = var_4357_end_0, end_mask = var_4357_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4357_cast_fp16")]; tensor var_4364_begin_0 = const()[name = string("op_4364_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4364_end_0 = const()[name = string("op_4364_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4364_end_mask_0 = const()[name = string("op_4364_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4364_cast_fp16 = slice_by_index(begin = var_4364_begin_0, end = var_4364_end_0, end_mask = var_4364_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4364_cast_fp16")]; tensor var_4371_begin_0 = const()[name = string("op_4371_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4371_end_0 = const()[name = string("op_4371_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4371_end_mask_0 = const()[name = string("op_4371_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4371_cast_fp16 = slice_by_index(begin = var_4371_begin_0, end = var_4371_end_0, end_mask = var_4371_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4371_cast_fp16")]; tensor var_4378_begin_0 = const()[name = string("op_4378_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4378_end_0 = const()[name = string("op_4378_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4378_end_mask_0 = const()[name = string("op_4378_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4378_cast_fp16 = slice_by_index(begin = var_4378_begin_0, end = var_4378_end_0, end_mask = var_4378_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4378_cast_fp16")]; tensor var_4385_begin_0 = const()[name = string("op_4385_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4385_end_0 = const()[name = string("op_4385_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4385_end_mask_0 = const()[name = string("op_4385_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4385_cast_fp16 = slice_by_index(begin = var_4385_begin_0, end = var_4385_end_0, end_mask = var_4385_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4385_cast_fp16")]; tensor var_4392_begin_0 = const()[name = string("op_4392_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4392_end_0 = const()[name = string("op_4392_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4392_end_mask_0 = const()[name = string("op_4392_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4392_cast_fp16 = slice_by_index(begin = var_4392_begin_0, end = var_4392_end_0, end_mask = var_4392_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4392_cast_fp16")]; tensor var_4399_begin_0 = const()[name = string("op_4399_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4399_end_0 = const()[name = string("op_4399_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4399_end_mask_0 = const()[name = string("op_4399_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4399_cast_fp16 = slice_by_index(begin = var_4399_begin_0, end = var_4399_end_0, end_mask = var_4399_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4399_cast_fp16")]; tensor var_4406_begin_0 = const()[name = string("op_4406_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4406_end_0 = const()[name = string("op_4406_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4406_end_mask_0 = const()[name = string("op_4406_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4406_cast_fp16 = slice_by_index(begin = var_4406_begin_0, end = var_4406_end_0, end_mask = var_4406_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4406_cast_fp16")]; tensor var_4413_begin_0 = const()[name = string("op_4413_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4413_end_0 = const()[name = string("op_4413_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4413_end_mask_0 = const()[name = string("op_4413_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4413_cast_fp16 = slice_by_index(begin = var_4413_begin_0, end = var_4413_end_0, end_mask = var_4413_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4413_cast_fp16")]; tensor var_4420_begin_0 = const()[name = string("op_4420_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4420_end_0 = const()[name = string("op_4420_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4420_end_mask_0 = const()[name = string("op_4420_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4420_cast_fp16 = slice_by_index(begin = var_4420_begin_0, end = var_4420_end_0, end_mask = var_4420_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4420_cast_fp16")]; tensor var_4427_begin_0 = const()[name = string("op_4427_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4427_end_0 = const()[name = string("op_4427_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4427_end_mask_0 = const()[name = string("op_4427_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4427_cast_fp16 = slice_by_index(begin = var_4427_begin_0, end = var_4427_end_0, end_mask = var_4427_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4427_cast_fp16")]; tensor var_4434_begin_0 = const()[name = string("op_4434_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4434_end_0 = const()[name = string("op_4434_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4434_end_mask_0 = const()[name = string("op_4434_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4434_cast_fp16 = slice_by_index(begin = var_4434_begin_0, end = var_4434_end_0, end_mask = var_4434_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4434_cast_fp16")]; tensor var_4441_begin_0 = const()[name = string("op_4441_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4441_end_0 = const()[name = string("op_4441_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4441_end_mask_0 = const()[name = string("op_4441_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4441_cast_fp16 = slice_by_index(begin = var_4441_begin_0, end = var_4441_end_0, end_mask = var_4441_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4441_cast_fp16")]; tensor var_4448_begin_0 = const()[name = string("op_4448_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4448_end_0 = const()[name = string("op_4448_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4448_end_mask_0 = const()[name = string("op_4448_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4448_cast_fp16 = slice_by_index(begin = var_4448_begin_0, end = var_4448_end_0, end_mask = var_4448_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4448_cast_fp16")]; tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_4453_begin_0 = const()[name = string("op_4453_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4453_end_0 = const()[name = string("op_4453_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_4453_end_mask_0 = const()[name = string("op_4453_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_7")]; tensor var_4453_cast_fp16 = slice_by_index(begin = var_4453_begin_0, end = var_4453_end_0, end_mask = var_4453_end_mask_0, x = k_9_cast_fp16)[name = string("op_4453_cast_fp16")]; tensor var_4457_begin_0 = const()[name = string("op_4457_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_4457_end_0 = const()[name = string("op_4457_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_4457_end_mask_0 = const()[name = string("op_4457_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4457_cast_fp16 = slice_by_index(begin = var_4457_begin_0, end = var_4457_end_0, end_mask = var_4457_end_mask_0, x = k_9_cast_fp16)[name = string("op_4457_cast_fp16")]; tensor var_4461_begin_0 = const()[name = string("op_4461_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_4461_end_0 = const()[name = string("op_4461_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_4461_end_mask_0 = const()[name = string("op_4461_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4461_cast_fp16 = slice_by_index(begin = var_4461_begin_0, end = var_4461_end_0, end_mask = var_4461_end_mask_0, x = k_9_cast_fp16)[name = string("op_4461_cast_fp16")]; tensor var_4465_begin_0 = const()[name = string("op_4465_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_4465_end_0 = const()[name = string("op_4465_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_4465_end_mask_0 = const()[name = string("op_4465_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4465_cast_fp16 = slice_by_index(begin = var_4465_begin_0, end = var_4465_end_0, end_mask = var_4465_end_mask_0, x = k_9_cast_fp16)[name = string("op_4465_cast_fp16")]; tensor var_4469_begin_0 = const()[name = string("op_4469_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_4469_end_0 = const()[name = string("op_4469_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_4469_end_mask_0 = const()[name = string("op_4469_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4469_cast_fp16 = slice_by_index(begin = var_4469_begin_0, end = var_4469_end_0, end_mask = var_4469_end_mask_0, x = k_9_cast_fp16)[name = string("op_4469_cast_fp16")]; tensor var_4473_begin_0 = const()[name = string("op_4473_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_4473_end_0 = const()[name = string("op_4473_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_4473_end_mask_0 = const()[name = string("op_4473_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4473_cast_fp16 = slice_by_index(begin = var_4473_begin_0, end = var_4473_end_0, end_mask = var_4473_end_mask_0, x = k_9_cast_fp16)[name = string("op_4473_cast_fp16")]; tensor var_4477_begin_0 = const()[name = string("op_4477_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_4477_end_0 = const()[name = string("op_4477_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_4477_end_mask_0 = const()[name = string("op_4477_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4477_cast_fp16 = slice_by_index(begin = var_4477_begin_0, end = var_4477_end_0, end_mask = var_4477_end_mask_0, x = k_9_cast_fp16)[name = string("op_4477_cast_fp16")]; tensor var_4481_begin_0 = const()[name = string("op_4481_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_4481_end_0 = const()[name = string("op_4481_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_4481_end_mask_0 = const()[name = string("op_4481_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4481_cast_fp16 = slice_by_index(begin = var_4481_begin_0, end = var_4481_end_0, end_mask = var_4481_end_mask_0, x = k_9_cast_fp16)[name = string("op_4481_cast_fp16")]; tensor var_4485_begin_0 = const()[name = string("op_4485_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4485_end_0 = const()[name = string("op_4485_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_4485_end_mask_0 = const()[name = string("op_4485_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4485_cast_fp16 = slice_by_index(begin = var_4485_begin_0, end = var_4485_end_0, end_mask = var_4485_end_mask_0, x = k_9_cast_fp16)[name = string("op_4485_cast_fp16")]; tensor var_4489_begin_0 = const()[name = string("op_4489_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_4489_end_0 = const()[name = string("op_4489_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_4489_end_mask_0 = const()[name = string("op_4489_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4489_cast_fp16 = slice_by_index(begin = var_4489_begin_0, end = var_4489_end_0, end_mask = var_4489_end_mask_0, x = k_9_cast_fp16)[name = string("op_4489_cast_fp16")]; tensor var_4493_begin_0 = const()[name = string("op_4493_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_4493_end_0 = const()[name = string("op_4493_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_4493_end_mask_0 = const()[name = string("op_4493_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4493_cast_fp16 = slice_by_index(begin = var_4493_begin_0, end = var_4493_end_0, end_mask = var_4493_end_mask_0, x = k_9_cast_fp16)[name = string("op_4493_cast_fp16")]; tensor var_4497_begin_0 = const()[name = string("op_4497_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_4497_end_0 = const()[name = string("op_4497_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_4497_end_mask_0 = const()[name = string("op_4497_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4497_cast_fp16 = slice_by_index(begin = var_4497_begin_0, end = var_4497_end_0, end_mask = var_4497_end_mask_0, x = k_9_cast_fp16)[name = string("op_4497_cast_fp16")]; tensor var_4499_begin_0 = const()[name = string("op_4499_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4499_end_0 = const()[name = string("op_4499_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4499_end_mask_0 = const()[name = string("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4499_cast_fp16 = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = value_9_cast_fp16)[name = string("op_4499_cast_fp16")]; tensor var_4503_begin_0 = const()[name = string("op_4503_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_4503_end_0 = const()[name = string("op_4503_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_4503_end_mask_0 = const()[name = string("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4503_cast_fp16 = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = value_9_cast_fp16)[name = string("op_4503_cast_fp16")]; tensor var_4507_begin_0 = const()[name = string("op_4507_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_4507_end_0 = const()[name = string("op_4507_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_4507_end_mask_0 = const()[name = string("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4507_cast_fp16 = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = value_9_cast_fp16)[name = string("op_4507_cast_fp16")]; tensor var_4511_begin_0 = const()[name = string("op_4511_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_4511_end_0 = const()[name = string("op_4511_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_4511_end_mask_0 = const()[name = string("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = value_9_cast_fp16)[name = string("op_4511_cast_fp16")]; tensor var_4515_begin_0 = const()[name = string("op_4515_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_4515_end_0 = const()[name = string("op_4515_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_4515_end_mask_0 = const()[name = string("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = value_9_cast_fp16)[name = string("op_4515_cast_fp16")]; tensor var_4519_begin_0 = const()[name = string("op_4519_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4519_end_0 = const()[name = string("op_4519_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_4519_end_mask_0 = const()[name = string("op_4519_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = value_9_cast_fp16)[name = string("op_4519_cast_fp16")]; tensor var_4523_begin_0 = const()[name = string("op_4523_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_4523_end_0 = const()[name = string("op_4523_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_4523_end_mask_0 = const()[name = string("op_4523_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4523_cast_fp16 = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = value_9_cast_fp16)[name = string("op_4523_cast_fp16")]; tensor var_4527_begin_0 = const()[name = string("op_4527_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_4527_end_0 = const()[name = string("op_4527_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_4527_end_mask_0 = const()[name = string("op_4527_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4527_cast_fp16 = slice_by_index(begin = var_4527_begin_0, end = var_4527_end_0, end_mask = var_4527_end_mask_0, x = value_9_cast_fp16)[name = string("op_4527_cast_fp16")]; tensor var_4531_begin_0 = const()[name = string("op_4531_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_4531_end_0 = const()[name = string("op_4531_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_4531_end_mask_0 = const()[name = string("op_4531_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4531_cast_fp16 = slice_by_index(begin = var_4531_begin_0, end = var_4531_end_0, end_mask = var_4531_end_mask_0, x = value_9_cast_fp16)[name = string("op_4531_cast_fp16")]; tensor var_4535_begin_0 = const()[name = string("op_4535_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_4535_end_0 = const()[name = string("op_4535_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_4535_end_mask_0 = const()[name = string("op_4535_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4535_cast_fp16 = slice_by_index(begin = var_4535_begin_0, end = var_4535_end_0, end_mask = var_4535_end_mask_0, x = value_9_cast_fp16)[name = string("op_4535_cast_fp16")]; tensor var_4539_begin_0 = const()[name = string("op_4539_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_4539_end_0 = const()[name = string("op_4539_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_4539_end_mask_0 = const()[name = string("op_4539_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4539_cast_fp16 = slice_by_index(begin = var_4539_begin_0, end = var_4539_end_0, end_mask = var_4539_end_mask_0, x = value_9_cast_fp16)[name = string("op_4539_cast_fp16")]; tensor var_4543_begin_0 = const()[name = string("op_4543_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_4543_end_0 = const()[name = string("op_4543_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_4543_end_mask_0 = const()[name = string("op_4543_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4543_cast_fp16 = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, x = value_9_cast_fp16)[name = string("op_4543_cast_fp16")]; string _SplitHeadsQ__mh_w_385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_385_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_385_equation_0, values = (var_4453_cast_fp16, var_4119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_385_cast_fp16")]; string _SplitHeadsQ__mh_w_387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_387_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_387_equation_0, values = (var_4453_cast_fp16, var_4126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_387_cast_fp16")]; string _SplitHeadsQ__mh_w_389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_389_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_389_equation_0, values = (var_4453_cast_fp16, var_4133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_389_cast_fp16")]; string _SplitHeadsQ__mh_w_391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_391_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_391_equation_0, values = (var_4453_cast_fp16, var_4140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_391_cast_fp16")]; string _SplitHeadsQ__mh_w_393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_393_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_393_equation_0, values = (var_4457_cast_fp16, var_4147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_393_cast_fp16")]; string _SplitHeadsQ__mh_w_395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_395_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_395_equation_0, values = (var_4457_cast_fp16, var_4154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_395_cast_fp16")]; string _SplitHeadsQ__mh_w_397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_397_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_397_equation_0, values = (var_4457_cast_fp16, var_4161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_397_cast_fp16")]; string _SplitHeadsQ__mh_w_399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_399_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_399_equation_0, values = (var_4457_cast_fp16, var_4168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_399_cast_fp16")]; string _SplitHeadsQ__mh_w_401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_401_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_401_equation_0, values = (var_4461_cast_fp16, var_4175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_401_cast_fp16")]; string _SplitHeadsQ__mh_w_403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_403_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_403_equation_0, values = (var_4461_cast_fp16, var_4182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_403_cast_fp16")]; string _SplitHeadsQ__mh_w_405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_405_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_405_equation_0, values = (var_4461_cast_fp16, var_4189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_405_cast_fp16")]; string _SplitHeadsQ__mh_w_407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_407_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_407_equation_0, values = (var_4461_cast_fp16, var_4196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_407_cast_fp16")]; string _SplitHeadsQ__mh_w_409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_409_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_409_equation_0, values = (var_4465_cast_fp16, var_4203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_409_cast_fp16")]; string _SplitHeadsQ__mh_w_411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_411_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_411_equation_0, values = (var_4465_cast_fp16, var_4210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_411_cast_fp16")]; string _SplitHeadsQ__mh_w_413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_413_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_413_equation_0, values = (var_4465_cast_fp16, var_4217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_413_cast_fp16")]; string _SplitHeadsQ__mh_w_415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_415_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_415_equation_0, values = (var_4465_cast_fp16, var_4224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_415_cast_fp16")]; string _SplitHeadsQ__mh_w_417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_417_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_417_equation_0, values = (var_4469_cast_fp16, var_4231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_417_cast_fp16")]; string _SplitHeadsQ__mh_w_419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_419_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_419_equation_0, values = (var_4469_cast_fp16, var_4238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_419_cast_fp16")]; string _SplitHeadsQ__mh_w_421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_421_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_421_equation_0, values = (var_4469_cast_fp16, var_4245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_421_cast_fp16")]; string _SplitHeadsQ__mh_w_423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_423_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_423_equation_0, values = (var_4469_cast_fp16, var_4252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_423_cast_fp16")]; string _SplitHeadsQ__mh_w_425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_425_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_425_equation_0, values = (var_4473_cast_fp16, var_4259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_425_cast_fp16")]; string _SplitHeadsQ__mh_w_427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_427_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_427_equation_0, values = (var_4473_cast_fp16, var_4266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_427_cast_fp16")]; string _SplitHeadsQ__mh_w_429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_429_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_429_equation_0, values = (var_4473_cast_fp16, var_4273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_429_cast_fp16")]; string _SplitHeadsQ__mh_w_431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_431_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_431_equation_0, values = (var_4473_cast_fp16, var_4280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_431_cast_fp16")]; string _SplitHeadsQ__mh_w_433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_433_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_433_equation_0, values = (var_4477_cast_fp16, var_4287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_433_cast_fp16")]; string _SplitHeadsQ__mh_w_435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_435_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_435_equation_0, values = (var_4477_cast_fp16, var_4294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_435_cast_fp16")]; string _SplitHeadsQ__mh_w_437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_437_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_437_equation_0, values = (var_4477_cast_fp16, var_4301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_437_cast_fp16")]; string _SplitHeadsQ__mh_w_439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_439_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_439_equation_0, values = (var_4477_cast_fp16, var_4308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_439_cast_fp16")]; string _SplitHeadsQ__mh_w_441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_441_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_441_equation_0, values = (var_4481_cast_fp16, var_4315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_441_cast_fp16")]; string _SplitHeadsQ__mh_w_443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_443_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_443_equation_0, values = (var_4481_cast_fp16, var_4322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_443_cast_fp16")]; string _SplitHeadsQ__mh_w_445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_445_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_445_equation_0, values = (var_4481_cast_fp16, var_4329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_445_cast_fp16")]; string _SplitHeadsQ__mh_w_447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_447_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_447_equation_0, values = (var_4481_cast_fp16, var_4336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_447_cast_fp16")]; string _SplitHeadsQ__mh_w_449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_449_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_449_equation_0, values = (var_4485_cast_fp16, var_4343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_449_cast_fp16")]; string _SplitHeadsQ__mh_w_451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_451_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_451_equation_0, values = (var_4485_cast_fp16, var_4350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_451_cast_fp16")]; string _SplitHeadsQ__mh_w_453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_453_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_453_equation_0, values = (var_4485_cast_fp16, var_4357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_453_cast_fp16")]; string _SplitHeadsQ__mh_w_455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_455_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_455_equation_0, values = (var_4485_cast_fp16, var_4364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_455_cast_fp16")]; string _SplitHeadsQ__mh_w_457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_457_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_457_equation_0, values = (var_4489_cast_fp16, var_4371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_457_cast_fp16")]; string _SplitHeadsQ__mh_w_459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_459_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_459_equation_0, values = (var_4489_cast_fp16, var_4378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_459_cast_fp16")]; string _SplitHeadsQ__mh_w_461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_461_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_461_equation_0, values = (var_4489_cast_fp16, var_4385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_461_cast_fp16")]; string _SplitHeadsQ__mh_w_463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_463_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_463_equation_0, values = (var_4489_cast_fp16, var_4392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_463_cast_fp16")]; string _SplitHeadsQ__mh_w_465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_465_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_465_equation_0, values = (var_4493_cast_fp16, var_4399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_465_cast_fp16")]; string _SplitHeadsQ__mh_w_467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_467_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_467_equation_0, values = (var_4493_cast_fp16, var_4406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_467_cast_fp16")]; string _SplitHeadsQ__mh_w_469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_469_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_469_equation_0, values = (var_4493_cast_fp16, var_4413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_469_cast_fp16")]; string _SplitHeadsQ__mh_w_471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_471_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_471_equation_0, values = (var_4493_cast_fp16, var_4420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_471_cast_fp16")]; string _SplitHeadsQ__mh_w_473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_473_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_473_equation_0, values = (var_4497_cast_fp16, var_4427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_473_cast_fp16")]; string _SplitHeadsQ__mh_w_475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_475_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_475_equation_0, values = (var_4497_cast_fp16, var_4434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_475_cast_fp16")]; string _SplitHeadsQ__mh_w_477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_477_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_477_equation_0, values = (var_4497_cast_fp16, var_4441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_477_cast_fp16")]; string _SplitHeadsQ__mh_w_479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_479_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_479_equation_0, values = (var_4497_cast_fp16, var_4448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_479_cast_fp16")]; fp16 var_4642_to_fp16 = const()[name = string("op_4642_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_385_cast_fp16, y = var_4642_to_fp16)[name = string("aw_chunk_385_cast_fp16")]; fp16 var_4644_to_fp16 = const()[name = string("op_4644_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_387_cast_fp16, y = var_4644_to_fp16)[name = string("aw_chunk_387_cast_fp16")]; fp16 var_4646_to_fp16 = const()[name = string("op_4646_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_389_cast_fp16, y = var_4646_to_fp16)[name = string("aw_chunk_389_cast_fp16")]; fp16 var_4648_to_fp16 = const()[name = string("op_4648_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_391_cast_fp16, y = var_4648_to_fp16)[name = string("aw_chunk_391_cast_fp16")]; fp16 var_4650_to_fp16 = const()[name = string("op_4650_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_393_cast_fp16, y = var_4650_to_fp16)[name = string("aw_chunk_393_cast_fp16")]; fp16 var_4652_to_fp16 = const()[name = string("op_4652_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_395_cast_fp16, y = var_4652_to_fp16)[name = string("aw_chunk_395_cast_fp16")]; fp16 var_4654_to_fp16 = const()[name = string("op_4654_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_397_cast_fp16, y = var_4654_to_fp16)[name = string("aw_chunk_397_cast_fp16")]; fp16 var_4656_to_fp16 = const()[name = string("op_4656_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_399_cast_fp16, y = var_4656_to_fp16)[name = string("aw_chunk_399_cast_fp16")]; fp16 var_4658_to_fp16 = const()[name = string("op_4658_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_401_cast_fp16, y = var_4658_to_fp16)[name = string("aw_chunk_401_cast_fp16")]; fp16 var_4660_to_fp16 = const()[name = string("op_4660_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_403_cast_fp16, y = var_4660_to_fp16)[name = string("aw_chunk_403_cast_fp16")]; fp16 var_4662_to_fp16 = const()[name = string("op_4662_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_405_cast_fp16, y = var_4662_to_fp16)[name = string("aw_chunk_405_cast_fp16")]; fp16 var_4664_to_fp16 = const()[name = string("op_4664_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_407_cast_fp16, y = var_4664_to_fp16)[name = string("aw_chunk_407_cast_fp16")]; fp16 var_4666_to_fp16 = const()[name = string("op_4666_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_409_cast_fp16, y = var_4666_to_fp16)[name = string("aw_chunk_409_cast_fp16")]; fp16 var_4668_to_fp16 = const()[name = string("op_4668_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_411_cast_fp16, y = var_4668_to_fp16)[name = string("aw_chunk_411_cast_fp16")]; fp16 var_4670_to_fp16 = const()[name = string("op_4670_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_413_cast_fp16, y = var_4670_to_fp16)[name = string("aw_chunk_413_cast_fp16")]; fp16 var_4672_to_fp16 = const()[name = string("op_4672_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_415_cast_fp16, y = var_4672_to_fp16)[name = string("aw_chunk_415_cast_fp16")]; fp16 var_4674_to_fp16 = const()[name = string("op_4674_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_417_cast_fp16, y = var_4674_to_fp16)[name = string("aw_chunk_417_cast_fp16")]; fp16 var_4676_to_fp16 = const()[name = string("op_4676_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_419_cast_fp16, y = var_4676_to_fp16)[name = string("aw_chunk_419_cast_fp16")]; fp16 var_4678_to_fp16 = const()[name = string("op_4678_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_421_cast_fp16, y = var_4678_to_fp16)[name = string("aw_chunk_421_cast_fp16")]; fp16 var_4680_to_fp16 = const()[name = string("op_4680_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_423_cast_fp16, y = var_4680_to_fp16)[name = string("aw_chunk_423_cast_fp16")]; fp16 var_4682_to_fp16 = const()[name = string("op_4682_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_425_cast_fp16, y = var_4682_to_fp16)[name = string("aw_chunk_425_cast_fp16")]; fp16 var_4684_to_fp16 = const()[name = string("op_4684_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_427_cast_fp16, y = var_4684_to_fp16)[name = string("aw_chunk_427_cast_fp16")]; fp16 var_4686_to_fp16 = const()[name = string("op_4686_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_429_cast_fp16, y = var_4686_to_fp16)[name = string("aw_chunk_429_cast_fp16")]; fp16 var_4688_to_fp16 = const()[name = string("op_4688_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_431_cast_fp16, y = var_4688_to_fp16)[name = string("aw_chunk_431_cast_fp16")]; fp16 var_4690_to_fp16 = const()[name = string("op_4690_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_433_cast_fp16, y = var_4690_to_fp16)[name = string("aw_chunk_433_cast_fp16")]; fp16 var_4692_to_fp16 = const()[name = string("op_4692_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_435_cast_fp16, y = var_4692_to_fp16)[name = string("aw_chunk_435_cast_fp16")]; fp16 var_4694_to_fp16 = const()[name = string("op_4694_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_437_cast_fp16, y = var_4694_to_fp16)[name = string("aw_chunk_437_cast_fp16")]; fp16 var_4696_to_fp16 = const()[name = string("op_4696_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_439_cast_fp16, y = var_4696_to_fp16)[name = string("aw_chunk_439_cast_fp16")]; fp16 var_4698_to_fp16 = const()[name = string("op_4698_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_441_cast_fp16, y = var_4698_to_fp16)[name = string("aw_chunk_441_cast_fp16")]; fp16 var_4700_to_fp16 = const()[name = string("op_4700_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_443_cast_fp16, y = var_4700_to_fp16)[name = string("aw_chunk_443_cast_fp16")]; fp16 var_4702_to_fp16 = const()[name = string("op_4702_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_445_cast_fp16, y = var_4702_to_fp16)[name = string("aw_chunk_445_cast_fp16")]; fp16 var_4704_to_fp16 = const()[name = string("op_4704_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_447_cast_fp16, y = var_4704_to_fp16)[name = string("aw_chunk_447_cast_fp16")]; fp16 var_4706_to_fp16 = const()[name = string("op_4706_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_449_cast_fp16, y = var_4706_to_fp16)[name = string("aw_chunk_449_cast_fp16")]; fp16 var_4708_to_fp16 = const()[name = string("op_4708_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_451_cast_fp16, y = var_4708_to_fp16)[name = string("aw_chunk_451_cast_fp16")]; fp16 var_4710_to_fp16 = const()[name = string("op_4710_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_453_cast_fp16, y = var_4710_to_fp16)[name = string("aw_chunk_453_cast_fp16")]; fp16 var_4712_to_fp16 = const()[name = string("op_4712_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_455_cast_fp16, y = var_4712_to_fp16)[name = string("aw_chunk_455_cast_fp16")]; fp16 var_4714_to_fp16 = const()[name = string("op_4714_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_457_cast_fp16, y = var_4714_to_fp16)[name = string("aw_chunk_457_cast_fp16")]; fp16 var_4716_to_fp16 = const()[name = string("op_4716_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_459_cast_fp16, y = var_4716_to_fp16)[name = string("aw_chunk_459_cast_fp16")]; fp16 var_4718_to_fp16 = const()[name = string("op_4718_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_461_cast_fp16, y = var_4718_to_fp16)[name = string("aw_chunk_461_cast_fp16")]; fp16 var_4720_to_fp16 = const()[name = string("op_4720_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_463_cast_fp16, y = var_4720_to_fp16)[name = string("aw_chunk_463_cast_fp16")]; fp16 var_4722_to_fp16 = const()[name = string("op_4722_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_465_cast_fp16, y = var_4722_to_fp16)[name = string("aw_chunk_465_cast_fp16")]; fp16 var_4724_to_fp16 = const()[name = string("op_4724_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_467_cast_fp16, y = var_4724_to_fp16)[name = string("aw_chunk_467_cast_fp16")]; fp16 var_4726_to_fp16 = const()[name = string("op_4726_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_469_cast_fp16, y = var_4726_to_fp16)[name = string("aw_chunk_469_cast_fp16")]; fp16 var_4728_to_fp16 = const()[name = string("op_4728_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_471_cast_fp16, y = var_4728_to_fp16)[name = string("aw_chunk_471_cast_fp16")]; fp16 var_4730_to_fp16 = const()[name = string("op_4730_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_473_cast_fp16, y = var_4730_to_fp16)[name = string("aw_chunk_473_cast_fp16")]; fp16 var_4732_to_fp16 = const()[name = string("op_4732_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_475_cast_fp16, y = var_4732_to_fp16)[name = string("aw_chunk_475_cast_fp16")]; fp16 var_4734_to_fp16 = const()[name = string("op_4734_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_477_cast_fp16, y = var_4734_to_fp16)[name = string("aw_chunk_477_cast_fp16")]; fp16 var_4736_to_fp16 = const()[name = string("op_4736_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_479_cast_fp16, y = var_4736_to_fp16)[name = string("aw_chunk_479_cast_fp16")]; tensor var_4738_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_385_cast_fp16)[name = string("op_4738_cast_fp16")]; tensor var_4739_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_387_cast_fp16)[name = string("op_4739_cast_fp16")]; tensor var_4740_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_389_cast_fp16)[name = string("op_4740_cast_fp16")]; tensor var_4741_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_391_cast_fp16)[name = string("op_4741_cast_fp16")]; tensor var_4742_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_393_cast_fp16)[name = string("op_4742_cast_fp16")]; tensor var_4743_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_395_cast_fp16)[name = string("op_4743_cast_fp16")]; tensor var_4744_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_397_cast_fp16)[name = string("op_4744_cast_fp16")]; tensor var_4745_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_399_cast_fp16)[name = string("op_4745_cast_fp16")]; tensor var_4746_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_401_cast_fp16)[name = string("op_4746_cast_fp16")]; tensor var_4747_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_403_cast_fp16)[name = string("op_4747_cast_fp16")]; tensor var_4748_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_405_cast_fp16)[name = string("op_4748_cast_fp16")]; tensor var_4749_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_407_cast_fp16)[name = string("op_4749_cast_fp16")]; tensor var_4750_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_409_cast_fp16)[name = string("op_4750_cast_fp16")]; tensor var_4751_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_411_cast_fp16)[name = string("op_4751_cast_fp16")]; tensor var_4752_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_413_cast_fp16)[name = string("op_4752_cast_fp16")]; tensor var_4753_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_415_cast_fp16)[name = string("op_4753_cast_fp16")]; tensor var_4754_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_417_cast_fp16)[name = string("op_4754_cast_fp16")]; tensor var_4755_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_419_cast_fp16)[name = string("op_4755_cast_fp16")]; tensor var_4756_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_421_cast_fp16)[name = string("op_4756_cast_fp16")]; tensor var_4757_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_423_cast_fp16)[name = string("op_4757_cast_fp16")]; tensor var_4758_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_425_cast_fp16)[name = string("op_4758_cast_fp16")]; tensor var_4759_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_427_cast_fp16)[name = string("op_4759_cast_fp16")]; tensor var_4760_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_429_cast_fp16)[name = string("op_4760_cast_fp16")]; tensor var_4761_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_431_cast_fp16)[name = string("op_4761_cast_fp16")]; tensor var_4762_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_433_cast_fp16)[name = string("op_4762_cast_fp16")]; tensor var_4763_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_435_cast_fp16)[name = string("op_4763_cast_fp16")]; tensor var_4764_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_437_cast_fp16)[name = string("op_4764_cast_fp16")]; tensor var_4765_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_439_cast_fp16)[name = string("op_4765_cast_fp16")]; tensor var_4766_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_441_cast_fp16)[name = string("op_4766_cast_fp16")]; tensor var_4767_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_443_cast_fp16)[name = string("op_4767_cast_fp16")]; tensor var_4768_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_445_cast_fp16)[name = string("op_4768_cast_fp16")]; tensor var_4769_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_447_cast_fp16)[name = string("op_4769_cast_fp16")]; tensor var_4770_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_449_cast_fp16)[name = string("op_4770_cast_fp16")]; tensor var_4771_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_451_cast_fp16)[name = string("op_4771_cast_fp16")]; tensor var_4772_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_453_cast_fp16)[name = string("op_4772_cast_fp16")]; tensor var_4773_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_455_cast_fp16)[name = string("op_4773_cast_fp16")]; tensor var_4774_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_457_cast_fp16)[name = string("op_4774_cast_fp16")]; tensor var_4775_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_459_cast_fp16)[name = string("op_4775_cast_fp16")]; tensor var_4776_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_461_cast_fp16)[name = string("op_4776_cast_fp16")]; tensor var_4777_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_463_cast_fp16)[name = string("op_4777_cast_fp16")]; tensor var_4778_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_465_cast_fp16)[name = string("op_4778_cast_fp16")]; tensor var_4779_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_467_cast_fp16)[name = string("op_4779_cast_fp16")]; tensor var_4780_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_469_cast_fp16)[name = string("op_4780_cast_fp16")]; tensor var_4781_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_471_cast_fp16)[name = string("op_4781_cast_fp16")]; tensor var_4782_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_473_cast_fp16)[name = string("op_4782_cast_fp16")]; tensor var_4783_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_475_cast_fp16)[name = string("op_4783_cast_fp16")]; tensor var_4784_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_477_cast_fp16)[name = string("op_4784_cast_fp16")]; tensor var_4785_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_479_cast_fp16)[name = string("op_4785_cast_fp16")]; string var_4787_equation_0 = const()[name = string("op_4787_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4787_cast_fp16 = einsum(equation = var_4787_equation_0, values = (var_4499_cast_fp16, var_4738_cast_fp16))[name = string("op_4787_cast_fp16")]; string var_4789_equation_0 = const()[name = string("op_4789_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4789_cast_fp16 = einsum(equation = var_4789_equation_0, values = (var_4499_cast_fp16, var_4739_cast_fp16))[name = string("op_4789_cast_fp16")]; string var_4791_equation_0 = const()[name = string("op_4791_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4791_cast_fp16 = einsum(equation = var_4791_equation_0, values = (var_4499_cast_fp16, var_4740_cast_fp16))[name = string("op_4791_cast_fp16")]; string var_4793_equation_0 = const()[name = string("op_4793_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4793_cast_fp16 = einsum(equation = var_4793_equation_0, values = (var_4499_cast_fp16, var_4741_cast_fp16))[name = string("op_4793_cast_fp16")]; string var_4795_equation_0 = const()[name = string("op_4795_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4795_cast_fp16 = einsum(equation = var_4795_equation_0, values = (var_4503_cast_fp16, var_4742_cast_fp16))[name = string("op_4795_cast_fp16")]; string var_4797_equation_0 = const()[name = string("op_4797_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4797_cast_fp16 = einsum(equation = var_4797_equation_0, values = (var_4503_cast_fp16, var_4743_cast_fp16))[name = string("op_4797_cast_fp16")]; string var_4799_equation_0 = const()[name = string("op_4799_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4799_cast_fp16 = einsum(equation = var_4799_equation_0, values = (var_4503_cast_fp16, var_4744_cast_fp16))[name = string("op_4799_cast_fp16")]; string var_4801_equation_0 = const()[name = string("op_4801_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4801_cast_fp16 = einsum(equation = var_4801_equation_0, values = (var_4503_cast_fp16, var_4745_cast_fp16))[name = string("op_4801_cast_fp16")]; string var_4803_equation_0 = const()[name = string("op_4803_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4803_cast_fp16 = einsum(equation = var_4803_equation_0, values = (var_4507_cast_fp16, var_4746_cast_fp16))[name = string("op_4803_cast_fp16")]; string var_4805_equation_0 = const()[name = string("op_4805_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4805_cast_fp16 = einsum(equation = var_4805_equation_0, values = (var_4507_cast_fp16, var_4747_cast_fp16))[name = string("op_4805_cast_fp16")]; string var_4807_equation_0 = const()[name = string("op_4807_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4807_cast_fp16 = einsum(equation = var_4807_equation_0, values = (var_4507_cast_fp16, var_4748_cast_fp16))[name = string("op_4807_cast_fp16")]; string var_4809_equation_0 = const()[name = string("op_4809_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4809_cast_fp16 = einsum(equation = var_4809_equation_0, values = (var_4507_cast_fp16, var_4749_cast_fp16))[name = string("op_4809_cast_fp16")]; string var_4811_equation_0 = const()[name = string("op_4811_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4811_cast_fp16 = einsum(equation = var_4811_equation_0, values = (var_4511_cast_fp16, var_4750_cast_fp16))[name = string("op_4811_cast_fp16")]; string var_4813_equation_0 = const()[name = string("op_4813_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4813_cast_fp16 = einsum(equation = var_4813_equation_0, values = (var_4511_cast_fp16, var_4751_cast_fp16))[name = string("op_4813_cast_fp16")]; string var_4815_equation_0 = const()[name = string("op_4815_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4815_cast_fp16 = einsum(equation = var_4815_equation_0, values = (var_4511_cast_fp16, var_4752_cast_fp16))[name = string("op_4815_cast_fp16")]; string var_4817_equation_0 = const()[name = string("op_4817_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4817_cast_fp16 = einsum(equation = var_4817_equation_0, values = (var_4511_cast_fp16, var_4753_cast_fp16))[name = string("op_4817_cast_fp16")]; string var_4819_equation_0 = const()[name = string("op_4819_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4819_cast_fp16 = einsum(equation = var_4819_equation_0, values = (var_4515_cast_fp16, var_4754_cast_fp16))[name = string("op_4819_cast_fp16")]; string var_4821_equation_0 = const()[name = string("op_4821_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4821_cast_fp16 = einsum(equation = var_4821_equation_0, values = (var_4515_cast_fp16, var_4755_cast_fp16))[name = string("op_4821_cast_fp16")]; string var_4823_equation_0 = const()[name = string("op_4823_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4823_cast_fp16 = einsum(equation = var_4823_equation_0, values = (var_4515_cast_fp16, var_4756_cast_fp16))[name = string("op_4823_cast_fp16")]; string var_4825_equation_0 = const()[name = string("op_4825_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4825_cast_fp16 = einsum(equation = var_4825_equation_0, values = (var_4515_cast_fp16, var_4757_cast_fp16))[name = string("op_4825_cast_fp16")]; string var_4827_equation_0 = const()[name = string("op_4827_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4827_cast_fp16 = einsum(equation = var_4827_equation_0, values = (var_4519_cast_fp16, var_4758_cast_fp16))[name = string("op_4827_cast_fp16")]; string var_4829_equation_0 = const()[name = string("op_4829_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4829_cast_fp16 = einsum(equation = var_4829_equation_0, values = (var_4519_cast_fp16, var_4759_cast_fp16))[name = string("op_4829_cast_fp16")]; string var_4831_equation_0 = const()[name = string("op_4831_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4519_cast_fp16, var_4760_cast_fp16))[name = string("op_4831_cast_fp16")]; string var_4833_equation_0 = const()[name = string("op_4833_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4833_cast_fp16 = einsum(equation = var_4833_equation_0, values = (var_4519_cast_fp16, var_4761_cast_fp16))[name = string("op_4833_cast_fp16")]; string var_4835_equation_0 = const()[name = string("op_4835_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4523_cast_fp16, var_4762_cast_fp16))[name = string("op_4835_cast_fp16")]; string var_4837_equation_0 = const()[name = string("op_4837_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4837_cast_fp16 = einsum(equation = var_4837_equation_0, values = (var_4523_cast_fp16, var_4763_cast_fp16))[name = string("op_4837_cast_fp16")]; string var_4839_equation_0 = const()[name = string("op_4839_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4523_cast_fp16, var_4764_cast_fp16))[name = string("op_4839_cast_fp16")]; string var_4841_equation_0 = const()[name = string("op_4841_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4841_cast_fp16 = einsum(equation = var_4841_equation_0, values = (var_4523_cast_fp16, var_4765_cast_fp16))[name = string("op_4841_cast_fp16")]; string var_4843_equation_0 = const()[name = string("op_4843_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4527_cast_fp16, var_4766_cast_fp16))[name = string("op_4843_cast_fp16")]; string var_4845_equation_0 = const()[name = string("op_4845_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4845_cast_fp16 = einsum(equation = var_4845_equation_0, values = (var_4527_cast_fp16, var_4767_cast_fp16))[name = string("op_4845_cast_fp16")]; string var_4847_equation_0 = const()[name = string("op_4847_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4527_cast_fp16, var_4768_cast_fp16))[name = string("op_4847_cast_fp16")]; string var_4849_equation_0 = const()[name = string("op_4849_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4849_cast_fp16 = einsum(equation = var_4849_equation_0, values = (var_4527_cast_fp16, var_4769_cast_fp16))[name = string("op_4849_cast_fp16")]; string var_4851_equation_0 = const()[name = string("op_4851_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4531_cast_fp16, var_4770_cast_fp16))[name = string("op_4851_cast_fp16")]; string var_4853_equation_0 = const()[name = string("op_4853_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4853_cast_fp16 = einsum(equation = var_4853_equation_0, values = (var_4531_cast_fp16, var_4771_cast_fp16))[name = string("op_4853_cast_fp16")]; string var_4855_equation_0 = const()[name = string("op_4855_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4531_cast_fp16, var_4772_cast_fp16))[name = string("op_4855_cast_fp16")]; string var_4857_equation_0 = const()[name = string("op_4857_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4857_cast_fp16 = einsum(equation = var_4857_equation_0, values = (var_4531_cast_fp16, var_4773_cast_fp16))[name = string("op_4857_cast_fp16")]; string var_4859_equation_0 = const()[name = string("op_4859_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4535_cast_fp16, var_4774_cast_fp16))[name = string("op_4859_cast_fp16")]; string var_4861_equation_0 = const()[name = string("op_4861_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4861_cast_fp16 = einsum(equation = var_4861_equation_0, values = (var_4535_cast_fp16, var_4775_cast_fp16))[name = string("op_4861_cast_fp16")]; string var_4863_equation_0 = const()[name = string("op_4863_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4535_cast_fp16, var_4776_cast_fp16))[name = string("op_4863_cast_fp16")]; string var_4865_equation_0 = const()[name = string("op_4865_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4865_cast_fp16 = einsum(equation = var_4865_equation_0, values = (var_4535_cast_fp16, var_4777_cast_fp16))[name = string("op_4865_cast_fp16")]; string var_4867_equation_0 = const()[name = string("op_4867_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4539_cast_fp16, var_4778_cast_fp16))[name = string("op_4867_cast_fp16")]; string var_4869_equation_0 = const()[name = string("op_4869_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4869_cast_fp16 = einsum(equation = var_4869_equation_0, values = (var_4539_cast_fp16, var_4779_cast_fp16))[name = string("op_4869_cast_fp16")]; string var_4871_equation_0 = const()[name = string("op_4871_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4539_cast_fp16, var_4780_cast_fp16))[name = string("op_4871_cast_fp16")]; string var_4873_equation_0 = const()[name = string("op_4873_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4873_cast_fp16 = einsum(equation = var_4873_equation_0, values = (var_4539_cast_fp16, var_4781_cast_fp16))[name = string("op_4873_cast_fp16")]; string var_4875_equation_0 = const()[name = string("op_4875_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4543_cast_fp16, var_4782_cast_fp16))[name = string("op_4875_cast_fp16")]; string var_4877_equation_0 = const()[name = string("op_4877_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4877_cast_fp16 = einsum(equation = var_4877_equation_0, values = (var_4543_cast_fp16, var_4783_cast_fp16))[name = string("op_4877_cast_fp16")]; string var_4879_equation_0 = const()[name = string("op_4879_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4543_cast_fp16, var_4784_cast_fp16))[name = string("op_4879_cast_fp16")]; string var_4881_equation_0 = const()[name = string("op_4881_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_4881_cast_fp16 = einsum(equation = var_4881_equation_0, values = (var_4543_cast_fp16, var_4785_cast_fp16))[name = string("op_4881_cast_fp16")]; bool var_4883_interleave_0 = const()[name = string("op_4883_interleave_0"), val = bool(false)]; tensor var_4883_cast_fp16 = concat(axis = var_3994, interleave = var_4883_interleave_0, values = (var_4787_cast_fp16, var_4789_cast_fp16, var_4791_cast_fp16, var_4793_cast_fp16))[name = string("op_4883_cast_fp16")]; bool var_4885_interleave_0 = const()[name = string("op_4885_interleave_0"), val = bool(false)]; tensor var_4885_cast_fp16 = concat(axis = var_3994, interleave = var_4885_interleave_0, values = (var_4795_cast_fp16, var_4797_cast_fp16, var_4799_cast_fp16, var_4801_cast_fp16))[name = string("op_4885_cast_fp16")]; bool var_4887_interleave_0 = const()[name = string("op_4887_interleave_0"), val = bool(false)]; tensor var_4887_cast_fp16 = concat(axis = var_3994, interleave = var_4887_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16))[name = string("op_4887_cast_fp16")]; bool var_4889_interleave_0 = const()[name = string("op_4889_interleave_0"), val = bool(false)]; tensor var_4889_cast_fp16 = concat(axis = var_3994, interleave = var_4889_interleave_0, values = (var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16))[name = string("op_4889_cast_fp16")]; bool var_4891_interleave_0 = const()[name = string("op_4891_interleave_0"), val = bool(false)]; tensor var_4891_cast_fp16 = concat(axis = var_3994, interleave = var_4891_interleave_0, values = (var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16))[name = string("op_4891_cast_fp16")]; bool var_4893_interleave_0 = const()[name = string("op_4893_interleave_0"), val = bool(false)]; tensor var_4893_cast_fp16 = concat(axis = var_3994, interleave = var_4893_interleave_0, values = (var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = string("op_4893_cast_fp16")]; bool var_4895_interleave_0 = const()[name = string("op_4895_interleave_0"), val = bool(false)]; tensor var_4895_cast_fp16 = concat(axis = var_3994, interleave = var_4895_interleave_0, values = (var_4835_cast_fp16, var_4837_cast_fp16, var_4839_cast_fp16, var_4841_cast_fp16))[name = string("op_4895_cast_fp16")]; bool var_4897_interleave_0 = const()[name = string("op_4897_interleave_0"), val = bool(false)]; tensor var_4897_cast_fp16 = concat(axis = var_3994, interleave = var_4897_interleave_0, values = (var_4843_cast_fp16, var_4845_cast_fp16, var_4847_cast_fp16, var_4849_cast_fp16))[name = string("op_4897_cast_fp16")]; bool var_4899_interleave_0 = const()[name = string("op_4899_interleave_0"), val = bool(false)]; tensor var_4899_cast_fp16 = concat(axis = var_3994, interleave = var_4899_interleave_0, values = (var_4851_cast_fp16, var_4853_cast_fp16, var_4855_cast_fp16, var_4857_cast_fp16))[name = string("op_4899_cast_fp16")]; bool var_4901_interleave_0 = const()[name = string("op_4901_interleave_0"), val = bool(false)]; tensor var_4901_cast_fp16 = concat(axis = var_3994, interleave = var_4901_interleave_0, values = (var_4859_cast_fp16, var_4861_cast_fp16, var_4863_cast_fp16, var_4865_cast_fp16))[name = string("op_4901_cast_fp16")]; bool var_4903_interleave_0 = const()[name = string("op_4903_interleave_0"), val = bool(false)]; tensor var_4903_cast_fp16 = concat(axis = var_3994, interleave = var_4903_interleave_0, values = (var_4867_cast_fp16, var_4869_cast_fp16, var_4871_cast_fp16, var_4873_cast_fp16))[name = string("op_4903_cast_fp16")]; bool var_4905_interleave_0 = const()[name = string("op_4905_interleave_0"), val = bool(false)]; tensor var_4905_cast_fp16 = concat(axis = var_3994, interleave = var_4905_interleave_0, values = (var_4875_cast_fp16, var_4877_cast_fp16, var_4879_cast_fp16, var_4881_cast_fp16))[name = string("op_4905_cast_fp16")]; bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; tensor input_33_cast_fp16 = concat(axis = var_4011, interleave = input_33_interleave_0, values = (var_4883_cast_fp16, var_4885_cast_fp16, var_4887_cast_fp16, var_4889_cast_fp16, var_4891_cast_fp16, var_4893_cast_fp16, var_4895_cast_fp16, var_4897_cast_fp16, var_4899_cast_fp16, var_4901_cast_fp16, var_4903_cast_fp16, var_4905_cast_fp16))[name = string("input_33_cast_fp16")]; string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66464448)))]; tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67644160)))]; tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; fp16 var_4924_to_fp16 = const()[name = string("op_4924_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_4924_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67645760)))]; tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67647360)))]; fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67648960)))]; tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72367616)))]; tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72373824)))]; tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77092480)))]; tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; int32 var_4953 = const()[name = string("op_4953"), val = int32(3)]; int32 var_4970 = const()[name = string("op_4970"), val = int32(1)]; tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; fp16 var_4987_to_fp16 = const()[name = string("op_4987_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_4987_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77094080)))]; tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77095680)))]; fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77097280)))]; tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78276992)))]; tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78278592)))]; tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79458304)))]; tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80638016)))]; tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; tensor var_5025_begin_0 = const()[name = string("op_5025_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5025_end_0 = const()[name = string("op_5025_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5025_end_mask_0 = const()[name = string("op_5025_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5025_cast_fp16 = slice_by_index(begin = var_5025_begin_0, end = var_5025_end_0, end_mask = var_5025_end_mask_0, x = query_11_cast_fp16)[name = string("op_5025_cast_fp16")]; tensor var_5029_begin_0 = const()[name = string("op_5029_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5029_end_0 = const()[name = string("op_5029_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5029_end_mask_0 = const()[name = string("op_5029_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5029_cast_fp16 = slice_by_index(begin = var_5029_begin_0, end = var_5029_end_0, end_mask = var_5029_end_mask_0, x = query_11_cast_fp16)[name = string("op_5029_cast_fp16")]; tensor var_5033_begin_0 = const()[name = string("op_5033_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5033_end_0 = const()[name = string("op_5033_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5033_end_mask_0 = const()[name = string("op_5033_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5033_cast_fp16 = slice_by_index(begin = var_5033_begin_0, end = var_5033_end_0, end_mask = var_5033_end_mask_0, x = query_11_cast_fp16)[name = string("op_5033_cast_fp16")]; tensor var_5037_begin_0 = const()[name = string("op_5037_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5037_end_0 = const()[name = string("op_5037_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5037_end_mask_0 = const()[name = string("op_5037_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5037_cast_fp16 = slice_by_index(begin = var_5037_begin_0, end = var_5037_end_0, end_mask = var_5037_end_mask_0, x = query_11_cast_fp16)[name = string("op_5037_cast_fp16")]; tensor var_5041_begin_0 = const()[name = string("op_5041_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5041_end_0 = const()[name = string("op_5041_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_5041_end_mask_0 = const()[name = string("op_5041_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5041_cast_fp16 = slice_by_index(begin = var_5041_begin_0, end = var_5041_end_0, end_mask = var_5041_end_mask_0, x = query_11_cast_fp16)[name = string("op_5041_cast_fp16")]; tensor var_5045_begin_0 = const()[name = string("op_5045_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5045_end_0 = const()[name = string("op_5045_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_5045_end_mask_0 = const()[name = string("op_5045_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5045_cast_fp16 = slice_by_index(begin = var_5045_begin_0, end = var_5045_end_0, end_mask = var_5045_end_mask_0, x = query_11_cast_fp16)[name = string("op_5045_cast_fp16")]; tensor var_5049_begin_0 = const()[name = string("op_5049_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5049_end_0 = const()[name = string("op_5049_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_5049_end_mask_0 = const()[name = string("op_5049_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5049_cast_fp16 = slice_by_index(begin = var_5049_begin_0, end = var_5049_end_0, end_mask = var_5049_end_mask_0, x = query_11_cast_fp16)[name = string("op_5049_cast_fp16")]; tensor var_5053_begin_0 = const()[name = string("op_5053_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5053_end_0 = const()[name = string("op_5053_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_5053_end_mask_0 = const()[name = string("op_5053_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5053_cast_fp16 = slice_by_index(begin = var_5053_begin_0, end = var_5053_end_0, end_mask = var_5053_end_mask_0, x = query_11_cast_fp16)[name = string("op_5053_cast_fp16")]; tensor var_5057_begin_0 = const()[name = string("op_5057_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5057_end_0 = const()[name = string("op_5057_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_5057_end_mask_0 = const()[name = string("op_5057_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5057_cast_fp16 = slice_by_index(begin = var_5057_begin_0, end = var_5057_end_0, end_mask = var_5057_end_mask_0, x = query_11_cast_fp16)[name = string("op_5057_cast_fp16")]; tensor var_5061_begin_0 = const()[name = string("op_5061_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5061_end_0 = const()[name = string("op_5061_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_5061_end_mask_0 = const()[name = string("op_5061_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5061_cast_fp16 = slice_by_index(begin = var_5061_begin_0, end = var_5061_end_0, end_mask = var_5061_end_mask_0, x = query_11_cast_fp16)[name = string("op_5061_cast_fp16")]; tensor var_5065_begin_0 = const()[name = string("op_5065_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5065_end_0 = const()[name = string("op_5065_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_5065_end_mask_0 = const()[name = string("op_5065_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5065_cast_fp16 = slice_by_index(begin = var_5065_begin_0, end = var_5065_end_0, end_mask = var_5065_end_mask_0, x = query_11_cast_fp16)[name = string("op_5065_cast_fp16")]; tensor var_5069_begin_0 = const()[name = string("op_5069_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5069_end_0 = const()[name = string("op_5069_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_5069_end_mask_0 = const()[name = string("op_5069_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5069_cast_fp16 = slice_by_index(begin = var_5069_begin_0, end = var_5069_end_0, end_mask = var_5069_end_mask_0, x = query_11_cast_fp16)[name = string("op_5069_cast_fp16")]; tensor var_5078_begin_0 = const()[name = string("op_5078_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5078_end_0 = const()[name = string("op_5078_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5078_end_mask_0 = const()[name = string("op_5078_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5078_cast_fp16")]; tensor var_5085_begin_0 = const()[name = string("op_5085_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5085_end_0 = const()[name = string("op_5085_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5085_end_mask_0 = const()[name = string("op_5085_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5085_cast_fp16 = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5085_cast_fp16")]; tensor var_5092_begin_0 = const()[name = string("op_5092_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5092_end_0 = const()[name = string("op_5092_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5092_end_mask_0 = const()[name = string("op_5092_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5092_cast_fp16 = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5092_cast_fp16")]; tensor var_5099_begin_0 = const()[name = string("op_5099_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5099_end_0 = const()[name = string("op_5099_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5099_end_mask_0 = const()[name = string("op_5099_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5099_cast_fp16 = slice_by_index(begin = var_5099_begin_0, end = var_5099_end_0, end_mask = var_5099_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5099_cast_fp16")]; tensor var_5106_begin_0 = const()[name = string("op_5106_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5106_end_0 = const()[name = string("op_5106_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5106_end_mask_0 = const()[name = string("op_5106_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5106_cast_fp16 = slice_by_index(begin = var_5106_begin_0, end = var_5106_end_0, end_mask = var_5106_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5106_cast_fp16")]; tensor var_5113_begin_0 = const()[name = string("op_5113_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5113_end_0 = const()[name = string("op_5113_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5113_end_mask_0 = const()[name = string("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5113_cast_fp16")]; tensor var_5120_begin_0 = const()[name = string("op_5120_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5120_end_0 = const()[name = string("op_5120_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5120_end_mask_0 = const()[name = string("op_5120_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5120_cast_fp16 = slice_by_index(begin = var_5120_begin_0, end = var_5120_end_0, end_mask = var_5120_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5120_cast_fp16")]; tensor var_5127_begin_0 = const()[name = string("op_5127_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5127_end_0 = const()[name = string("op_5127_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5127_end_mask_0 = const()[name = string("op_5127_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5127_cast_fp16 = slice_by_index(begin = var_5127_begin_0, end = var_5127_end_0, end_mask = var_5127_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5127_cast_fp16")]; tensor var_5134_begin_0 = const()[name = string("op_5134_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5134_end_0 = const()[name = string("op_5134_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5134_end_mask_0 = const()[name = string("op_5134_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5134_cast_fp16 = slice_by_index(begin = var_5134_begin_0, end = var_5134_end_0, end_mask = var_5134_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5134_cast_fp16")]; tensor var_5141_begin_0 = const()[name = string("op_5141_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5141_end_0 = const()[name = string("op_5141_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5141_end_mask_0 = const()[name = string("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5141_cast_fp16")]; tensor var_5148_begin_0 = const()[name = string("op_5148_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5148_end_0 = const()[name = string("op_5148_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5148_end_mask_0 = const()[name = string("op_5148_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5148_cast_fp16 = slice_by_index(begin = var_5148_begin_0, end = var_5148_end_0, end_mask = var_5148_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5148_cast_fp16")]; tensor var_5155_begin_0 = const()[name = string("op_5155_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5155_end_0 = const()[name = string("op_5155_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5155_end_mask_0 = const()[name = string("op_5155_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5155_cast_fp16 = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5155_cast_fp16")]; tensor var_5162_begin_0 = const()[name = string("op_5162_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5162_end_0 = const()[name = string("op_5162_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5162_end_mask_0 = const()[name = string("op_5162_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5162_cast_fp16 = slice_by_index(begin = var_5162_begin_0, end = var_5162_end_0, end_mask = var_5162_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5162_cast_fp16")]; tensor var_5169_begin_0 = const()[name = string("op_5169_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5169_end_0 = const()[name = string("op_5169_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5169_end_mask_0 = const()[name = string("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5169_cast_fp16")]; tensor var_5176_begin_0 = const()[name = string("op_5176_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5176_end_0 = const()[name = string("op_5176_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5176_end_mask_0 = const()[name = string("op_5176_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5176_cast_fp16 = slice_by_index(begin = var_5176_begin_0, end = var_5176_end_0, end_mask = var_5176_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5176_cast_fp16")]; tensor var_5183_begin_0 = const()[name = string("op_5183_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5183_end_0 = const()[name = string("op_5183_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5183_end_mask_0 = const()[name = string("op_5183_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5183_cast_fp16")]; tensor var_5190_begin_0 = const()[name = string("op_5190_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5190_end_0 = const()[name = string("op_5190_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5190_end_mask_0 = const()[name = string("op_5190_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5190_cast_fp16 = slice_by_index(begin = var_5190_begin_0, end = var_5190_end_0, end_mask = var_5190_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5190_cast_fp16")]; tensor var_5197_begin_0 = const()[name = string("op_5197_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5197_end_0 = const()[name = string("op_5197_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5197_end_mask_0 = const()[name = string("op_5197_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5197_cast_fp16 = slice_by_index(begin = var_5197_begin_0, end = var_5197_end_0, end_mask = var_5197_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5197_cast_fp16")]; tensor var_5204_begin_0 = const()[name = string("op_5204_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5204_end_0 = const()[name = string("op_5204_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5204_end_mask_0 = const()[name = string("op_5204_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5204_cast_fp16 = slice_by_index(begin = var_5204_begin_0, end = var_5204_end_0, end_mask = var_5204_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5204_cast_fp16")]; tensor var_5211_begin_0 = const()[name = string("op_5211_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5211_end_0 = const()[name = string("op_5211_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5211_end_mask_0 = const()[name = string("op_5211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5211_cast_fp16")]; tensor var_5218_begin_0 = const()[name = string("op_5218_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5218_end_0 = const()[name = string("op_5218_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5218_end_mask_0 = const()[name = string("op_5218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5218_cast_fp16 = slice_by_index(begin = var_5218_begin_0, end = var_5218_end_0, end_mask = var_5218_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5218_cast_fp16")]; tensor var_5225_begin_0 = const()[name = string("op_5225_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5225_end_0 = const()[name = string("op_5225_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5225_end_mask_0 = const()[name = string("op_5225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5225_cast_fp16 = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5225_cast_fp16")]; tensor var_5232_begin_0 = const()[name = string("op_5232_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5232_end_0 = const()[name = string("op_5232_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5232_end_mask_0 = const()[name = string("op_5232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5232_cast_fp16")]; tensor var_5239_begin_0 = const()[name = string("op_5239_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5239_end_0 = const()[name = string("op_5239_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5239_end_mask_0 = const()[name = string("op_5239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5239_cast_fp16")]; tensor var_5246_begin_0 = const()[name = string("op_5246_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5246_end_0 = const()[name = string("op_5246_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5246_end_mask_0 = const()[name = string("op_5246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5246_cast_fp16 = slice_by_index(begin = var_5246_begin_0, end = var_5246_end_0, end_mask = var_5246_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5246_cast_fp16")]; tensor var_5253_begin_0 = const()[name = string("op_5253_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5253_end_0 = const()[name = string("op_5253_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5253_end_mask_0 = const()[name = string("op_5253_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5253_cast_fp16 = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5253_cast_fp16")]; tensor var_5260_begin_0 = const()[name = string("op_5260_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5260_end_0 = const()[name = string("op_5260_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5260_end_mask_0 = const()[name = string("op_5260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5260_cast_fp16 = slice_by_index(begin = var_5260_begin_0, end = var_5260_end_0, end_mask = var_5260_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5260_cast_fp16")]; tensor var_5267_begin_0 = const()[name = string("op_5267_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5267_end_0 = const()[name = string("op_5267_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5267_end_mask_0 = const()[name = string("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5267_cast_fp16")]; tensor var_5274_begin_0 = const()[name = string("op_5274_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5274_end_0 = const()[name = string("op_5274_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5274_end_mask_0 = const()[name = string("op_5274_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5274_cast_fp16 = slice_by_index(begin = var_5274_begin_0, end = var_5274_end_0, end_mask = var_5274_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5274_cast_fp16")]; tensor var_5281_begin_0 = const()[name = string("op_5281_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5281_end_0 = const()[name = string("op_5281_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5281_end_mask_0 = const()[name = string("op_5281_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5281_cast_fp16")]; tensor var_5288_begin_0 = const()[name = string("op_5288_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5288_end_0 = const()[name = string("op_5288_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5288_end_mask_0 = const()[name = string("op_5288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5288_cast_fp16 = slice_by_index(begin = var_5288_begin_0, end = var_5288_end_0, end_mask = var_5288_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5288_cast_fp16")]; tensor var_5295_begin_0 = const()[name = string("op_5295_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5295_end_0 = const()[name = string("op_5295_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5295_end_mask_0 = const()[name = string("op_5295_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5295_cast_fp16 = slice_by_index(begin = var_5295_begin_0, end = var_5295_end_0, end_mask = var_5295_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5295_cast_fp16")]; tensor var_5302_begin_0 = const()[name = string("op_5302_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5302_end_0 = const()[name = string("op_5302_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5302_end_mask_0 = const()[name = string("op_5302_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5302_cast_fp16 = slice_by_index(begin = var_5302_begin_0, end = var_5302_end_0, end_mask = var_5302_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5302_cast_fp16")]; tensor var_5309_begin_0 = const()[name = string("op_5309_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5309_end_0 = const()[name = string("op_5309_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5309_end_mask_0 = const()[name = string("op_5309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5309_cast_fp16 = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5309_cast_fp16")]; tensor var_5316_begin_0 = const()[name = string("op_5316_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5316_end_0 = const()[name = string("op_5316_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5316_end_mask_0 = const()[name = string("op_5316_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5316_cast_fp16 = slice_by_index(begin = var_5316_begin_0, end = var_5316_end_0, end_mask = var_5316_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5316_cast_fp16")]; tensor var_5323_begin_0 = const()[name = string("op_5323_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5323_end_0 = const()[name = string("op_5323_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5323_end_mask_0 = const()[name = string("op_5323_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5323_cast_fp16 = slice_by_index(begin = var_5323_begin_0, end = var_5323_end_0, end_mask = var_5323_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5323_cast_fp16")]; tensor var_5330_begin_0 = const()[name = string("op_5330_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5330_end_0 = const()[name = string("op_5330_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5330_end_mask_0 = const()[name = string("op_5330_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5330_cast_fp16 = slice_by_index(begin = var_5330_begin_0, end = var_5330_end_0, end_mask = var_5330_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5330_cast_fp16")]; tensor var_5337_begin_0 = const()[name = string("op_5337_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5337_end_0 = const()[name = string("op_5337_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5337_end_mask_0 = const()[name = string("op_5337_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5337_cast_fp16 = slice_by_index(begin = var_5337_begin_0, end = var_5337_end_0, end_mask = var_5337_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5337_cast_fp16")]; tensor var_5344_begin_0 = const()[name = string("op_5344_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5344_end_0 = const()[name = string("op_5344_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5344_end_mask_0 = const()[name = string("op_5344_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5344_cast_fp16 = slice_by_index(begin = var_5344_begin_0, end = var_5344_end_0, end_mask = var_5344_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5344_cast_fp16")]; tensor var_5351_begin_0 = const()[name = string("op_5351_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5351_end_0 = const()[name = string("op_5351_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5351_end_mask_0 = const()[name = string("op_5351_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5351_cast_fp16 = slice_by_index(begin = var_5351_begin_0, end = var_5351_end_0, end_mask = var_5351_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5351_cast_fp16")]; tensor var_5358_begin_0 = const()[name = string("op_5358_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5358_end_0 = const()[name = string("op_5358_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5358_end_mask_0 = const()[name = string("op_5358_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5358_cast_fp16 = slice_by_index(begin = var_5358_begin_0, end = var_5358_end_0, end_mask = var_5358_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5358_cast_fp16")]; tensor var_5365_begin_0 = const()[name = string("op_5365_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5365_end_0 = const()[name = string("op_5365_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5365_end_mask_0 = const()[name = string("op_5365_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5365_cast_fp16 = slice_by_index(begin = var_5365_begin_0, end = var_5365_end_0, end_mask = var_5365_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5365_cast_fp16")]; tensor var_5372_begin_0 = const()[name = string("op_5372_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5372_end_0 = const()[name = string("op_5372_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5372_end_mask_0 = const()[name = string("op_5372_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5372_cast_fp16 = slice_by_index(begin = var_5372_begin_0, end = var_5372_end_0, end_mask = var_5372_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5372_cast_fp16")]; tensor var_5379_begin_0 = const()[name = string("op_5379_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5379_end_0 = const()[name = string("op_5379_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5379_end_mask_0 = const()[name = string("op_5379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5379_cast_fp16 = slice_by_index(begin = var_5379_begin_0, end = var_5379_end_0, end_mask = var_5379_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5379_cast_fp16")]; tensor var_5386_begin_0 = const()[name = string("op_5386_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5386_end_0 = const()[name = string("op_5386_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5386_end_mask_0 = const()[name = string("op_5386_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5386_cast_fp16 = slice_by_index(begin = var_5386_begin_0, end = var_5386_end_0, end_mask = var_5386_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5386_cast_fp16")]; tensor var_5393_begin_0 = const()[name = string("op_5393_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5393_end_0 = const()[name = string("op_5393_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5393_end_mask_0 = const()[name = string("op_5393_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5393_cast_fp16 = slice_by_index(begin = var_5393_begin_0, end = var_5393_end_0, end_mask = var_5393_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5393_cast_fp16")]; tensor var_5400_begin_0 = const()[name = string("op_5400_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5400_end_0 = const()[name = string("op_5400_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5400_end_mask_0 = const()[name = string("op_5400_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5400_cast_fp16 = slice_by_index(begin = var_5400_begin_0, end = var_5400_end_0, end_mask = var_5400_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5400_cast_fp16")]; tensor var_5407_begin_0 = const()[name = string("op_5407_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5407_end_0 = const()[name = string("op_5407_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5407_end_mask_0 = const()[name = string("op_5407_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5407_cast_fp16 = slice_by_index(begin = var_5407_begin_0, end = var_5407_end_0, end_mask = var_5407_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5407_cast_fp16")]; tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_5412_begin_0 = const()[name = string("op_5412_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5412_end_0 = const()[name = string("op_5412_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_5412_end_mask_0 = const()[name = string("op_5412_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = string("transpose_6")]; tensor var_5412_cast_fp16 = slice_by_index(begin = var_5412_begin_0, end = var_5412_end_0, end_mask = var_5412_end_mask_0, x = k_11_cast_fp16)[name = string("op_5412_cast_fp16")]; tensor var_5416_begin_0 = const()[name = string("op_5416_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_5416_end_0 = const()[name = string("op_5416_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_5416_end_mask_0 = const()[name = string("op_5416_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5416_cast_fp16 = slice_by_index(begin = var_5416_begin_0, end = var_5416_end_0, end_mask = var_5416_end_mask_0, x = k_11_cast_fp16)[name = string("op_5416_cast_fp16")]; tensor var_5420_begin_0 = const()[name = string("op_5420_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_5420_end_0 = const()[name = string("op_5420_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_5420_end_mask_0 = const()[name = string("op_5420_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5420_cast_fp16 = slice_by_index(begin = var_5420_begin_0, end = var_5420_end_0, end_mask = var_5420_end_mask_0, x = k_11_cast_fp16)[name = string("op_5420_cast_fp16")]; tensor var_5424_begin_0 = const()[name = string("op_5424_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_5424_end_0 = const()[name = string("op_5424_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_5424_end_mask_0 = const()[name = string("op_5424_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5424_cast_fp16 = slice_by_index(begin = var_5424_begin_0, end = var_5424_end_0, end_mask = var_5424_end_mask_0, x = k_11_cast_fp16)[name = string("op_5424_cast_fp16")]; tensor var_5428_begin_0 = const()[name = string("op_5428_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_5428_end_0 = const()[name = string("op_5428_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_5428_end_mask_0 = const()[name = string("op_5428_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = k_11_cast_fp16)[name = string("op_5428_cast_fp16")]; tensor var_5432_begin_0 = const()[name = string("op_5432_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_5432_end_0 = const()[name = string("op_5432_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_5432_end_mask_0 = const()[name = string("op_5432_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = k_11_cast_fp16)[name = string("op_5432_cast_fp16")]; tensor var_5436_begin_0 = const()[name = string("op_5436_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_5436_end_0 = const()[name = string("op_5436_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_5436_end_mask_0 = const()[name = string("op_5436_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5436_cast_fp16 = slice_by_index(begin = var_5436_begin_0, end = var_5436_end_0, end_mask = var_5436_end_mask_0, x = k_11_cast_fp16)[name = string("op_5436_cast_fp16")]; tensor var_5440_begin_0 = const()[name = string("op_5440_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_5440_end_0 = const()[name = string("op_5440_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_5440_end_mask_0 = const()[name = string("op_5440_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5440_cast_fp16 = slice_by_index(begin = var_5440_begin_0, end = var_5440_end_0, end_mask = var_5440_end_mask_0, x = k_11_cast_fp16)[name = string("op_5440_cast_fp16")]; tensor var_5444_begin_0 = const()[name = string("op_5444_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5444_end_0 = const()[name = string("op_5444_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_5444_end_mask_0 = const()[name = string("op_5444_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5444_cast_fp16 = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = k_11_cast_fp16)[name = string("op_5444_cast_fp16")]; tensor var_5448_begin_0 = const()[name = string("op_5448_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_5448_end_0 = const()[name = string("op_5448_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_5448_end_mask_0 = const()[name = string("op_5448_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5448_cast_fp16 = slice_by_index(begin = var_5448_begin_0, end = var_5448_end_0, end_mask = var_5448_end_mask_0, x = k_11_cast_fp16)[name = string("op_5448_cast_fp16")]; tensor var_5452_begin_0 = const()[name = string("op_5452_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_5452_end_0 = const()[name = string("op_5452_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_5452_end_mask_0 = const()[name = string("op_5452_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5452_cast_fp16 = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = k_11_cast_fp16)[name = string("op_5452_cast_fp16")]; tensor var_5456_begin_0 = const()[name = string("op_5456_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_5456_end_0 = const()[name = string("op_5456_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_5456_end_mask_0 = const()[name = string("op_5456_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5456_cast_fp16 = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = k_11_cast_fp16)[name = string("op_5456_cast_fp16")]; tensor var_5458_begin_0 = const()[name = string("op_5458_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5458_end_0 = const()[name = string("op_5458_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5458_end_mask_0 = const()[name = string("op_5458_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5458_cast_fp16 = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = value_11_cast_fp16)[name = string("op_5458_cast_fp16")]; tensor var_5462_begin_0 = const()[name = string("op_5462_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5462_end_0 = const()[name = string("op_5462_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5462_end_mask_0 = const()[name = string("op_5462_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = value_11_cast_fp16)[name = string("op_5462_cast_fp16")]; tensor var_5466_begin_0 = const()[name = string("op_5466_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5466_end_0 = const()[name = string("op_5466_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5466_end_mask_0 = const()[name = string("op_5466_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5466_cast_fp16 = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = value_11_cast_fp16)[name = string("op_5466_cast_fp16")]; tensor var_5470_begin_0 = const()[name = string("op_5470_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5470_end_0 = const()[name = string("op_5470_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5470_end_mask_0 = const()[name = string("op_5470_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5470_cast_fp16 = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = value_11_cast_fp16)[name = string("op_5470_cast_fp16")]; tensor var_5474_begin_0 = const()[name = string("op_5474_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5474_end_0 = const()[name = string("op_5474_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_5474_end_mask_0 = const()[name = string("op_5474_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5474_cast_fp16 = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = value_11_cast_fp16)[name = string("op_5474_cast_fp16")]; tensor var_5478_begin_0 = const()[name = string("op_5478_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5478_end_0 = const()[name = string("op_5478_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_5478_end_mask_0 = const()[name = string("op_5478_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5478_cast_fp16 = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = value_11_cast_fp16)[name = string("op_5478_cast_fp16")]; tensor var_5482_begin_0 = const()[name = string("op_5482_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5482_end_0 = const()[name = string("op_5482_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_5482_end_mask_0 = const()[name = string("op_5482_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5482_cast_fp16 = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = value_11_cast_fp16)[name = string("op_5482_cast_fp16")]; tensor var_5486_begin_0 = const()[name = string("op_5486_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5486_end_0 = const()[name = string("op_5486_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_5486_end_mask_0 = const()[name = string("op_5486_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5486_cast_fp16 = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = value_11_cast_fp16)[name = string("op_5486_cast_fp16")]; tensor var_5490_begin_0 = const()[name = string("op_5490_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5490_end_0 = const()[name = string("op_5490_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_5490_end_mask_0 = const()[name = string("op_5490_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5490_cast_fp16 = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = value_11_cast_fp16)[name = string("op_5490_cast_fp16")]; tensor var_5494_begin_0 = const()[name = string("op_5494_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5494_end_0 = const()[name = string("op_5494_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_5494_end_mask_0 = const()[name = string("op_5494_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = value_11_cast_fp16)[name = string("op_5494_cast_fp16")]; tensor var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = value_11_cast_fp16)[name = string("op_5498_cast_fp16")]; tensor var_5502_begin_0 = const()[name = string("op_5502_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5502_end_0 = const()[name = string("op_5502_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_5502_end_mask_0 = const()[name = string("op_5502_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5502_cast_fp16 = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = value_11_cast_fp16)[name = string("op_5502_cast_fp16")]; string _SplitHeadsQ__mh_w_481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_481_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_481_equation_0, values = (var_5412_cast_fp16, var_5078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_481_cast_fp16")]; string _SplitHeadsQ__mh_w_483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_483_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_483_equation_0, values = (var_5412_cast_fp16, var_5085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_483_cast_fp16")]; string _SplitHeadsQ__mh_w_485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_485_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_485_equation_0, values = (var_5412_cast_fp16, var_5092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_485_cast_fp16")]; string _SplitHeadsQ__mh_w_487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_487_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_487_equation_0, values = (var_5412_cast_fp16, var_5099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_487_cast_fp16")]; string _SplitHeadsQ__mh_w_489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_489_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_489_equation_0, values = (var_5416_cast_fp16, var_5106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_489_cast_fp16")]; string _SplitHeadsQ__mh_w_491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_491_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_491_equation_0, values = (var_5416_cast_fp16, var_5113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_491_cast_fp16")]; string _SplitHeadsQ__mh_w_493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_493_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_493_equation_0, values = (var_5416_cast_fp16, var_5120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_493_cast_fp16")]; string _SplitHeadsQ__mh_w_495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_495_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_495_equation_0, values = (var_5416_cast_fp16, var_5127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_495_cast_fp16")]; string _SplitHeadsQ__mh_w_497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_497_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_497_equation_0, values = (var_5420_cast_fp16, var_5134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_497_cast_fp16")]; string _SplitHeadsQ__mh_w_499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_499_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_499_equation_0, values = (var_5420_cast_fp16, var_5141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_499_cast_fp16")]; string _SplitHeadsQ__mh_w_501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_501_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_501_equation_0, values = (var_5420_cast_fp16, var_5148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_501_cast_fp16")]; string _SplitHeadsQ__mh_w_503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_503_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_503_equation_0, values = (var_5420_cast_fp16, var_5155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_503_cast_fp16")]; string _SplitHeadsQ__mh_w_505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_505_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_505_equation_0, values = (var_5424_cast_fp16, var_5162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_505_cast_fp16")]; string _SplitHeadsQ__mh_w_507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_507_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_507_equation_0, values = (var_5424_cast_fp16, var_5169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_507_cast_fp16")]; string _SplitHeadsQ__mh_w_509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_509_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_509_equation_0, values = (var_5424_cast_fp16, var_5176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_509_cast_fp16")]; string _SplitHeadsQ__mh_w_511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_511_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_511_equation_0, values = (var_5424_cast_fp16, var_5183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_511_cast_fp16")]; string _SplitHeadsQ__mh_w_513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_513_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_513_equation_0, values = (var_5428_cast_fp16, var_5190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_513_cast_fp16")]; string _SplitHeadsQ__mh_w_515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_515_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_515_equation_0, values = (var_5428_cast_fp16, var_5197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_515_cast_fp16")]; string _SplitHeadsQ__mh_w_517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_517_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_517_equation_0, values = (var_5428_cast_fp16, var_5204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_517_cast_fp16")]; string _SplitHeadsQ__mh_w_519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_519_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_519_equation_0, values = (var_5428_cast_fp16, var_5211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_519_cast_fp16")]; string _SplitHeadsQ__mh_w_521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_521_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_521_equation_0, values = (var_5432_cast_fp16, var_5218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_521_cast_fp16")]; string _SplitHeadsQ__mh_w_523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_523_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_523_equation_0, values = (var_5432_cast_fp16, var_5225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_523_cast_fp16")]; string _SplitHeadsQ__mh_w_525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_525_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_525_equation_0, values = (var_5432_cast_fp16, var_5232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_525_cast_fp16")]; string _SplitHeadsQ__mh_w_527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_527_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_527_equation_0, values = (var_5432_cast_fp16, var_5239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_527_cast_fp16")]; string _SplitHeadsQ__mh_w_529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_529_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_529_equation_0, values = (var_5436_cast_fp16, var_5246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_529_cast_fp16")]; string _SplitHeadsQ__mh_w_531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_531_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_531_equation_0, values = (var_5436_cast_fp16, var_5253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_531_cast_fp16")]; string _SplitHeadsQ__mh_w_533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_533_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_533_equation_0, values = (var_5436_cast_fp16, var_5260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_533_cast_fp16")]; string _SplitHeadsQ__mh_w_535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_535_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_535_equation_0, values = (var_5436_cast_fp16, var_5267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_535_cast_fp16")]; string _SplitHeadsQ__mh_w_537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_537_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_537_equation_0, values = (var_5440_cast_fp16, var_5274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_537_cast_fp16")]; string _SplitHeadsQ__mh_w_539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_539_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_539_equation_0, values = (var_5440_cast_fp16, var_5281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_539_cast_fp16")]; string _SplitHeadsQ__mh_w_541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_541_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_541_equation_0, values = (var_5440_cast_fp16, var_5288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_541_cast_fp16")]; string _SplitHeadsQ__mh_w_543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_543_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_543_equation_0, values = (var_5440_cast_fp16, var_5295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_543_cast_fp16")]; string _SplitHeadsQ__mh_w_545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_545_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_545_equation_0, values = (var_5444_cast_fp16, var_5302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_545_cast_fp16")]; string _SplitHeadsQ__mh_w_547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_547_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_547_equation_0, values = (var_5444_cast_fp16, var_5309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_547_cast_fp16")]; string _SplitHeadsQ__mh_w_549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_549_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_549_equation_0, values = (var_5444_cast_fp16, var_5316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_549_cast_fp16")]; string _SplitHeadsQ__mh_w_551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_551_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_551_equation_0, values = (var_5444_cast_fp16, var_5323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_551_cast_fp16")]; string _SplitHeadsQ__mh_w_553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_553_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_553_equation_0, values = (var_5448_cast_fp16, var_5330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_553_cast_fp16")]; string _SplitHeadsQ__mh_w_555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_555_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_555_equation_0, values = (var_5448_cast_fp16, var_5337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_555_cast_fp16")]; string _SplitHeadsQ__mh_w_557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_557_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_557_equation_0, values = (var_5448_cast_fp16, var_5344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_557_cast_fp16")]; string _SplitHeadsQ__mh_w_559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_559_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_559_equation_0, values = (var_5448_cast_fp16, var_5351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_559_cast_fp16")]; string _SplitHeadsQ__mh_w_561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_561_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_561_equation_0, values = (var_5452_cast_fp16, var_5358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_561_cast_fp16")]; string _SplitHeadsQ__mh_w_563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_563_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_563_equation_0, values = (var_5452_cast_fp16, var_5365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_563_cast_fp16")]; string _SplitHeadsQ__mh_w_565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_565_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_565_equation_0, values = (var_5452_cast_fp16, var_5372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_565_cast_fp16")]; string _SplitHeadsQ__mh_w_567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_567_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_567_equation_0, values = (var_5452_cast_fp16, var_5379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_567_cast_fp16")]; string _SplitHeadsQ__mh_w_569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_569_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_569_equation_0, values = (var_5456_cast_fp16, var_5386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_569_cast_fp16")]; string _SplitHeadsQ__mh_w_571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_571_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_571_equation_0, values = (var_5456_cast_fp16, var_5393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_571_cast_fp16")]; string _SplitHeadsQ__mh_w_573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_573_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_573_equation_0, values = (var_5456_cast_fp16, var_5400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_573_cast_fp16")]; string _SplitHeadsQ__mh_w_575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_575_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_575_equation_0, values = (var_5456_cast_fp16, var_5407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_575_cast_fp16")]; fp16 var_5601_to_fp16 = const()[name = string("op_5601_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_481_cast_fp16, y = var_5601_to_fp16)[name = string("aw_chunk_481_cast_fp16")]; fp16 var_5603_to_fp16 = const()[name = string("op_5603_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_483_cast_fp16, y = var_5603_to_fp16)[name = string("aw_chunk_483_cast_fp16")]; fp16 var_5605_to_fp16 = const()[name = string("op_5605_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_485_cast_fp16, y = var_5605_to_fp16)[name = string("aw_chunk_485_cast_fp16")]; fp16 var_5607_to_fp16 = const()[name = string("op_5607_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_487_cast_fp16, y = var_5607_to_fp16)[name = string("aw_chunk_487_cast_fp16")]; fp16 var_5609_to_fp16 = const()[name = string("op_5609_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_489_cast_fp16, y = var_5609_to_fp16)[name = string("aw_chunk_489_cast_fp16")]; fp16 var_5611_to_fp16 = const()[name = string("op_5611_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_491_cast_fp16, y = var_5611_to_fp16)[name = string("aw_chunk_491_cast_fp16")]; fp16 var_5613_to_fp16 = const()[name = string("op_5613_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_493_cast_fp16, y = var_5613_to_fp16)[name = string("aw_chunk_493_cast_fp16")]; fp16 var_5615_to_fp16 = const()[name = string("op_5615_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_495_cast_fp16, y = var_5615_to_fp16)[name = string("aw_chunk_495_cast_fp16")]; fp16 var_5617_to_fp16 = const()[name = string("op_5617_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_497_cast_fp16, y = var_5617_to_fp16)[name = string("aw_chunk_497_cast_fp16")]; fp16 var_5619_to_fp16 = const()[name = string("op_5619_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_499_cast_fp16, y = var_5619_to_fp16)[name = string("aw_chunk_499_cast_fp16")]; fp16 var_5621_to_fp16 = const()[name = string("op_5621_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_501_cast_fp16, y = var_5621_to_fp16)[name = string("aw_chunk_501_cast_fp16")]; fp16 var_5623_to_fp16 = const()[name = string("op_5623_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_503_cast_fp16, y = var_5623_to_fp16)[name = string("aw_chunk_503_cast_fp16")]; fp16 var_5625_to_fp16 = const()[name = string("op_5625_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_505_cast_fp16, y = var_5625_to_fp16)[name = string("aw_chunk_505_cast_fp16")]; fp16 var_5627_to_fp16 = const()[name = string("op_5627_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_507_cast_fp16, y = var_5627_to_fp16)[name = string("aw_chunk_507_cast_fp16")]; fp16 var_5629_to_fp16 = const()[name = string("op_5629_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_509_cast_fp16, y = var_5629_to_fp16)[name = string("aw_chunk_509_cast_fp16")]; fp16 var_5631_to_fp16 = const()[name = string("op_5631_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_511_cast_fp16, y = var_5631_to_fp16)[name = string("aw_chunk_511_cast_fp16")]; fp16 var_5633_to_fp16 = const()[name = string("op_5633_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_513_cast_fp16, y = var_5633_to_fp16)[name = string("aw_chunk_513_cast_fp16")]; fp16 var_5635_to_fp16 = const()[name = string("op_5635_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_515_cast_fp16, y = var_5635_to_fp16)[name = string("aw_chunk_515_cast_fp16")]; fp16 var_5637_to_fp16 = const()[name = string("op_5637_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_517_cast_fp16, y = var_5637_to_fp16)[name = string("aw_chunk_517_cast_fp16")]; fp16 var_5639_to_fp16 = const()[name = string("op_5639_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_519_cast_fp16, y = var_5639_to_fp16)[name = string("aw_chunk_519_cast_fp16")]; fp16 var_5641_to_fp16 = const()[name = string("op_5641_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_521_cast_fp16, y = var_5641_to_fp16)[name = string("aw_chunk_521_cast_fp16")]; fp16 var_5643_to_fp16 = const()[name = string("op_5643_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_523_cast_fp16, y = var_5643_to_fp16)[name = string("aw_chunk_523_cast_fp16")]; fp16 var_5645_to_fp16 = const()[name = string("op_5645_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_525_cast_fp16, y = var_5645_to_fp16)[name = string("aw_chunk_525_cast_fp16")]; fp16 var_5647_to_fp16 = const()[name = string("op_5647_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_527_cast_fp16, y = var_5647_to_fp16)[name = string("aw_chunk_527_cast_fp16")]; fp16 var_5649_to_fp16 = const()[name = string("op_5649_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_529_cast_fp16, y = var_5649_to_fp16)[name = string("aw_chunk_529_cast_fp16")]; fp16 var_5651_to_fp16 = const()[name = string("op_5651_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_531_cast_fp16, y = var_5651_to_fp16)[name = string("aw_chunk_531_cast_fp16")]; fp16 var_5653_to_fp16 = const()[name = string("op_5653_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_533_cast_fp16, y = var_5653_to_fp16)[name = string("aw_chunk_533_cast_fp16")]; fp16 var_5655_to_fp16 = const()[name = string("op_5655_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_535_cast_fp16, y = var_5655_to_fp16)[name = string("aw_chunk_535_cast_fp16")]; fp16 var_5657_to_fp16 = const()[name = string("op_5657_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_537_cast_fp16, y = var_5657_to_fp16)[name = string("aw_chunk_537_cast_fp16")]; fp16 var_5659_to_fp16 = const()[name = string("op_5659_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_539_cast_fp16, y = var_5659_to_fp16)[name = string("aw_chunk_539_cast_fp16")]; fp16 var_5661_to_fp16 = const()[name = string("op_5661_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_541_cast_fp16, y = var_5661_to_fp16)[name = string("aw_chunk_541_cast_fp16")]; fp16 var_5663_to_fp16 = const()[name = string("op_5663_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_543_cast_fp16, y = var_5663_to_fp16)[name = string("aw_chunk_543_cast_fp16")]; fp16 var_5665_to_fp16 = const()[name = string("op_5665_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_545_cast_fp16, y = var_5665_to_fp16)[name = string("aw_chunk_545_cast_fp16")]; fp16 var_5667_to_fp16 = const()[name = string("op_5667_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_547_cast_fp16, y = var_5667_to_fp16)[name = string("aw_chunk_547_cast_fp16")]; fp16 var_5669_to_fp16 = const()[name = string("op_5669_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_549_cast_fp16, y = var_5669_to_fp16)[name = string("aw_chunk_549_cast_fp16")]; fp16 var_5671_to_fp16 = const()[name = string("op_5671_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_551_cast_fp16, y = var_5671_to_fp16)[name = string("aw_chunk_551_cast_fp16")]; fp16 var_5673_to_fp16 = const()[name = string("op_5673_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_553_cast_fp16, y = var_5673_to_fp16)[name = string("aw_chunk_553_cast_fp16")]; fp16 var_5675_to_fp16 = const()[name = string("op_5675_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_555_cast_fp16, y = var_5675_to_fp16)[name = string("aw_chunk_555_cast_fp16")]; fp16 var_5677_to_fp16 = const()[name = string("op_5677_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_557_cast_fp16, y = var_5677_to_fp16)[name = string("aw_chunk_557_cast_fp16")]; fp16 var_5679_to_fp16 = const()[name = string("op_5679_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_559_cast_fp16, y = var_5679_to_fp16)[name = string("aw_chunk_559_cast_fp16")]; fp16 var_5681_to_fp16 = const()[name = string("op_5681_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_561_cast_fp16, y = var_5681_to_fp16)[name = string("aw_chunk_561_cast_fp16")]; fp16 var_5683_to_fp16 = const()[name = string("op_5683_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_563_cast_fp16, y = var_5683_to_fp16)[name = string("aw_chunk_563_cast_fp16")]; fp16 var_5685_to_fp16 = const()[name = string("op_5685_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_565_cast_fp16, y = var_5685_to_fp16)[name = string("aw_chunk_565_cast_fp16")]; fp16 var_5687_to_fp16 = const()[name = string("op_5687_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_567_cast_fp16, y = var_5687_to_fp16)[name = string("aw_chunk_567_cast_fp16")]; fp16 var_5689_to_fp16 = const()[name = string("op_5689_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_569_cast_fp16, y = var_5689_to_fp16)[name = string("aw_chunk_569_cast_fp16")]; fp16 var_5691_to_fp16 = const()[name = string("op_5691_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_571_cast_fp16, y = var_5691_to_fp16)[name = string("aw_chunk_571_cast_fp16")]; fp16 var_5693_to_fp16 = const()[name = string("op_5693_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_573_cast_fp16, y = var_5693_to_fp16)[name = string("aw_chunk_573_cast_fp16")]; fp16 var_5695_to_fp16 = const()[name = string("op_5695_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_575_cast_fp16, y = var_5695_to_fp16)[name = string("aw_chunk_575_cast_fp16")]; tensor var_5697_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_481_cast_fp16)[name = string("op_5697_cast_fp16")]; tensor var_5698_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_483_cast_fp16)[name = string("op_5698_cast_fp16")]; tensor var_5699_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_485_cast_fp16)[name = string("op_5699_cast_fp16")]; tensor var_5700_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_487_cast_fp16)[name = string("op_5700_cast_fp16")]; tensor var_5701_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_489_cast_fp16)[name = string("op_5701_cast_fp16")]; tensor var_5702_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_491_cast_fp16)[name = string("op_5702_cast_fp16")]; tensor var_5703_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_493_cast_fp16)[name = string("op_5703_cast_fp16")]; tensor var_5704_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_495_cast_fp16)[name = string("op_5704_cast_fp16")]; tensor var_5705_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_497_cast_fp16)[name = string("op_5705_cast_fp16")]; tensor var_5706_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_499_cast_fp16)[name = string("op_5706_cast_fp16")]; tensor var_5707_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_501_cast_fp16)[name = string("op_5707_cast_fp16")]; tensor var_5708_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_503_cast_fp16)[name = string("op_5708_cast_fp16")]; tensor var_5709_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_505_cast_fp16)[name = string("op_5709_cast_fp16")]; tensor var_5710_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_507_cast_fp16)[name = string("op_5710_cast_fp16")]; tensor var_5711_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_509_cast_fp16)[name = string("op_5711_cast_fp16")]; tensor var_5712_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_511_cast_fp16)[name = string("op_5712_cast_fp16")]; tensor var_5713_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_513_cast_fp16)[name = string("op_5713_cast_fp16")]; tensor var_5714_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_515_cast_fp16)[name = string("op_5714_cast_fp16")]; tensor var_5715_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_517_cast_fp16)[name = string("op_5715_cast_fp16")]; tensor var_5716_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_519_cast_fp16)[name = string("op_5716_cast_fp16")]; tensor var_5717_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_521_cast_fp16)[name = string("op_5717_cast_fp16")]; tensor var_5718_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_523_cast_fp16)[name = string("op_5718_cast_fp16")]; tensor var_5719_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_525_cast_fp16)[name = string("op_5719_cast_fp16")]; tensor var_5720_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_527_cast_fp16)[name = string("op_5720_cast_fp16")]; tensor var_5721_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_529_cast_fp16)[name = string("op_5721_cast_fp16")]; tensor var_5722_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_531_cast_fp16)[name = string("op_5722_cast_fp16")]; tensor var_5723_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_533_cast_fp16)[name = string("op_5723_cast_fp16")]; tensor var_5724_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_535_cast_fp16)[name = string("op_5724_cast_fp16")]; tensor var_5725_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_537_cast_fp16)[name = string("op_5725_cast_fp16")]; tensor var_5726_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_539_cast_fp16)[name = string("op_5726_cast_fp16")]; tensor var_5727_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_541_cast_fp16)[name = string("op_5727_cast_fp16")]; tensor var_5728_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_543_cast_fp16)[name = string("op_5728_cast_fp16")]; tensor var_5729_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_545_cast_fp16)[name = string("op_5729_cast_fp16")]; tensor var_5730_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_547_cast_fp16)[name = string("op_5730_cast_fp16")]; tensor var_5731_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_549_cast_fp16)[name = string("op_5731_cast_fp16")]; tensor var_5732_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_551_cast_fp16)[name = string("op_5732_cast_fp16")]; tensor var_5733_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_553_cast_fp16)[name = string("op_5733_cast_fp16")]; tensor var_5734_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_555_cast_fp16)[name = string("op_5734_cast_fp16")]; tensor var_5735_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_557_cast_fp16)[name = string("op_5735_cast_fp16")]; tensor var_5736_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_559_cast_fp16)[name = string("op_5736_cast_fp16")]; tensor var_5737_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_561_cast_fp16)[name = string("op_5737_cast_fp16")]; tensor var_5738_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_563_cast_fp16)[name = string("op_5738_cast_fp16")]; tensor var_5739_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_565_cast_fp16)[name = string("op_5739_cast_fp16")]; tensor var_5740_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_567_cast_fp16)[name = string("op_5740_cast_fp16")]; tensor var_5741_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_569_cast_fp16)[name = string("op_5741_cast_fp16")]; tensor var_5742_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_571_cast_fp16)[name = string("op_5742_cast_fp16")]; tensor var_5743_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_573_cast_fp16)[name = string("op_5743_cast_fp16")]; tensor var_5744_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_575_cast_fp16)[name = string("op_5744_cast_fp16")]; string var_5746_equation_0 = const()[name = string("op_5746_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5458_cast_fp16, var_5697_cast_fp16))[name = string("op_5746_cast_fp16")]; string var_5748_equation_0 = const()[name = string("op_5748_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5748_cast_fp16 = einsum(equation = var_5748_equation_0, values = (var_5458_cast_fp16, var_5698_cast_fp16))[name = string("op_5748_cast_fp16")]; string var_5750_equation_0 = const()[name = string("op_5750_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5458_cast_fp16, var_5699_cast_fp16))[name = string("op_5750_cast_fp16")]; string var_5752_equation_0 = const()[name = string("op_5752_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5752_cast_fp16 = einsum(equation = var_5752_equation_0, values = (var_5458_cast_fp16, var_5700_cast_fp16))[name = string("op_5752_cast_fp16")]; string var_5754_equation_0 = const()[name = string("op_5754_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5462_cast_fp16, var_5701_cast_fp16))[name = string("op_5754_cast_fp16")]; string var_5756_equation_0 = const()[name = string("op_5756_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5756_cast_fp16 = einsum(equation = var_5756_equation_0, values = (var_5462_cast_fp16, var_5702_cast_fp16))[name = string("op_5756_cast_fp16")]; string var_5758_equation_0 = const()[name = string("op_5758_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5462_cast_fp16, var_5703_cast_fp16))[name = string("op_5758_cast_fp16")]; string var_5760_equation_0 = const()[name = string("op_5760_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5760_cast_fp16 = einsum(equation = var_5760_equation_0, values = (var_5462_cast_fp16, var_5704_cast_fp16))[name = string("op_5760_cast_fp16")]; string var_5762_equation_0 = const()[name = string("op_5762_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5466_cast_fp16, var_5705_cast_fp16))[name = string("op_5762_cast_fp16")]; string var_5764_equation_0 = const()[name = string("op_5764_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5764_cast_fp16 = einsum(equation = var_5764_equation_0, values = (var_5466_cast_fp16, var_5706_cast_fp16))[name = string("op_5764_cast_fp16")]; string var_5766_equation_0 = const()[name = string("op_5766_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5466_cast_fp16, var_5707_cast_fp16))[name = string("op_5766_cast_fp16")]; string var_5768_equation_0 = const()[name = string("op_5768_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5768_cast_fp16 = einsum(equation = var_5768_equation_0, values = (var_5466_cast_fp16, var_5708_cast_fp16))[name = string("op_5768_cast_fp16")]; string var_5770_equation_0 = const()[name = string("op_5770_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5470_cast_fp16, var_5709_cast_fp16))[name = string("op_5770_cast_fp16")]; string var_5772_equation_0 = const()[name = string("op_5772_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5772_cast_fp16 = einsum(equation = var_5772_equation_0, values = (var_5470_cast_fp16, var_5710_cast_fp16))[name = string("op_5772_cast_fp16")]; string var_5774_equation_0 = const()[name = string("op_5774_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5470_cast_fp16, var_5711_cast_fp16))[name = string("op_5774_cast_fp16")]; string var_5776_equation_0 = const()[name = string("op_5776_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5776_cast_fp16 = einsum(equation = var_5776_equation_0, values = (var_5470_cast_fp16, var_5712_cast_fp16))[name = string("op_5776_cast_fp16")]; string var_5778_equation_0 = const()[name = string("op_5778_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5474_cast_fp16, var_5713_cast_fp16))[name = string("op_5778_cast_fp16")]; string var_5780_equation_0 = const()[name = string("op_5780_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5780_cast_fp16 = einsum(equation = var_5780_equation_0, values = (var_5474_cast_fp16, var_5714_cast_fp16))[name = string("op_5780_cast_fp16")]; string var_5782_equation_0 = const()[name = string("op_5782_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5474_cast_fp16, var_5715_cast_fp16))[name = string("op_5782_cast_fp16")]; string var_5784_equation_0 = const()[name = string("op_5784_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5784_cast_fp16 = einsum(equation = var_5784_equation_0, values = (var_5474_cast_fp16, var_5716_cast_fp16))[name = string("op_5784_cast_fp16")]; string var_5786_equation_0 = const()[name = string("op_5786_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5478_cast_fp16, var_5717_cast_fp16))[name = string("op_5786_cast_fp16")]; string var_5788_equation_0 = const()[name = string("op_5788_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5788_cast_fp16 = einsum(equation = var_5788_equation_0, values = (var_5478_cast_fp16, var_5718_cast_fp16))[name = string("op_5788_cast_fp16")]; string var_5790_equation_0 = const()[name = string("op_5790_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5478_cast_fp16, var_5719_cast_fp16))[name = string("op_5790_cast_fp16")]; string var_5792_equation_0 = const()[name = string("op_5792_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5792_cast_fp16 = einsum(equation = var_5792_equation_0, values = (var_5478_cast_fp16, var_5720_cast_fp16))[name = string("op_5792_cast_fp16")]; string var_5794_equation_0 = const()[name = string("op_5794_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5482_cast_fp16, var_5721_cast_fp16))[name = string("op_5794_cast_fp16")]; string var_5796_equation_0 = const()[name = string("op_5796_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5796_cast_fp16 = einsum(equation = var_5796_equation_0, values = (var_5482_cast_fp16, var_5722_cast_fp16))[name = string("op_5796_cast_fp16")]; string var_5798_equation_0 = const()[name = string("op_5798_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5482_cast_fp16, var_5723_cast_fp16))[name = string("op_5798_cast_fp16")]; string var_5800_equation_0 = const()[name = string("op_5800_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5800_cast_fp16 = einsum(equation = var_5800_equation_0, values = (var_5482_cast_fp16, var_5724_cast_fp16))[name = string("op_5800_cast_fp16")]; string var_5802_equation_0 = const()[name = string("op_5802_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5486_cast_fp16, var_5725_cast_fp16))[name = string("op_5802_cast_fp16")]; string var_5804_equation_0 = const()[name = string("op_5804_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5804_cast_fp16 = einsum(equation = var_5804_equation_0, values = (var_5486_cast_fp16, var_5726_cast_fp16))[name = string("op_5804_cast_fp16")]; string var_5806_equation_0 = const()[name = string("op_5806_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5486_cast_fp16, var_5727_cast_fp16))[name = string("op_5806_cast_fp16")]; string var_5808_equation_0 = const()[name = string("op_5808_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5808_cast_fp16 = einsum(equation = var_5808_equation_0, values = (var_5486_cast_fp16, var_5728_cast_fp16))[name = string("op_5808_cast_fp16")]; string var_5810_equation_0 = const()[name = string("op_5810_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5490_cast_fp16, var_5729_cast_fp16))[name = string("op_5810_cast_fp16")]; string var_5812_equation_0 = const()[name = string("op_5812_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5812_cast_fp16 = einsum(equation = var_5812_equation_0, values = (var_5490_cast_fp16, var_5730_cast_fp16))[name = string("op_5812_cast_fp16")]; string var_5814_equation_0 = const()[name = string("op_5814_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5490_cast_fp16, var_5731_cast_fp16))[name = string("op_5814_cast_fp16")]; string var_5816_equation_0 = const()[name = string("op_5816_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5816_cast_fp16 = einsum(equation = var_5816_equation_0, values = (var_5490_cast_fp16, var_5732_cast_fp16))[name = string("op_5816_cast_fp16")]; string var_5818_equation_0 = const()[name = string("op_5818_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5818_cast_fp16 = einsum(equation = var_5818_equation_0, values = (var_5494_cast_fp16, var_5733_cast_fp16))[name = string("op_5818_cast_fp16")]; string var_5820_equation_0 = const()[name = string("op_5820_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5820_cast_fp16 = einsum(equation = var_5820_equation_0, values = (var_5494_cast_fp16, var_5734_cast_fp16))[name = string("op_5820_cast_fp16")]; string var_5822_equation_0 = const()[name = string("op_5822_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5822_cast_fp16 = einsum(equation = var_5822_equation_0, values = (var_5494_cast_fp16, var_5735_cast_fp16))[name = string("op_5822_cast_fp16")]; string var_5824_equation_0 = const()[name = string("op_5824_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5824_cast_fp16 = einsum(equation = var_5824_equation_0, values = (var_5494_cast_fp16, var_5736_cast_fp16))[name = string("op_5824_cast_fp16")]; string var_5826_equation_0 = const()[name = string("op_5826_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5826_cast_fp16 = einsum(equation = var_5826_equation_0, values = (var_5498_cast_fp16, var_5737_cast_fp16))[name = string("op_5826_cast_fp16")]; string var_5828_equation_0 = const()[name = string("op_5828_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5828_cast_fp16 = einsum(equation = var_5828_equation_0, values = (var_5498_cast_fp16, var_5738_cast_fp16))[name = string("op_5828_cast_fp16")]; string var_5830_equation_0 = const()[name = string("op_5830_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5830_cast_fp16 = einsum(equation = var_5830_equation_0, values = (var_5498_cast_fp16, var_5739_cast_fp16))[name = string("op_5830_cast_fp16")]; string var_5832_equation_0 = const()[name = string("op_5832_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5832_cast_fp16 = einsum(equation = var_5832_equation_0, values = (var_5498_cast_fp16, var_5740_cast_fp16))[name = string("op_5832_cast_fp16")]; string var_5834_equation_0 = const()[name = string("op_5834_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5834_cast_fp16 = einsum(equation = var_5834_equation_0, values = (var_5502_cast_fp16, var_5741_cast_fp16))[name = string("op_5834_cast_fp16")]; string var_5836_equation_0 = const()[name = string("op_5836_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5836_cast_fp16 = einsum(equation = var_5836_equation_0, values = (var_5502_cast_fp16, var_5742_cast_fp16))[name = string("op_5836_cast_fp16")]; string var_5838_equation_0 = const()[name = string("op_5838_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5838_cast_fp16 = einsum(equation = var_5838_equation_0, values = (var_5502_cast_fp16, var_5743_cast_fp16))[name = string("op_5838_cast_fp16")]; string var_5840_equation_0 = const()[name = string("op_5840_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_5840_cast_fp16 = einsum(equation = var_5840_equation_0, values = (var_5502_cast_fp16, var_5744_cast_fp16))[name = string("op_5840_cast_fp16")]; bool var_5842_interleave_0 = const()[name = string("op_5842_interleave_0"), val = bool(false)]; tensor var_5842_cast_fp16 = concat(axis = var_4953, interleave = var_5842_interleave_0, values = (var_5746_cast_fp16, var_5748_cast_fp16, var_5750_cast_fp16, var_5752_cast_fp16))[name = string("op_5842_cast_fp16")]; bool var_5844_interleave_0 = const()[name = string("op_5844_interleave_0"), val = bool(false)]; tensor var_5844_cast_fp16 = concat(axis = var_4953, interleave = var_5844_interleave_0, values = (var_5754_cast_fp16, var_5756_cast_fp16, var_5758_cast_fp16, var_5760_cast_fp16))[name = string("op_5844_cast_fp16")]; bool var_5846_interleave_0 = const()[name = string("op_5846_interleave_0"), val = bool(false)]; tensor var_5846_cast_fp16 = concat(axis = var_4953, interleave = var_5846_interleave_0, values = (var_5762_cast_fp16, var_5764_cast_fp16, var_5766_cast_fp16, var_5768_cast_fp16))[name = string("op_5846_cast_fp16")]; bool var_5848_interleave_0 = const()[name = string("op_5848_interleave_0"), val = bool(false)]; tensor var_5848_cast_fp16 = concat(axis = var_4953, interleave = var_5848_interleave_0, values = (var_5770_cast_fp16, var_5772_cast_fp16, var_5774_cast_fp16, var_5776_cast_fp16))[name = string("op_5848_cast_fp16")]; bool var_5850_interleave_0 = const()[name = string("op_5850_interleave_0"), val = bool(false)]; tensor var_5850_cast_fp16 = concat(axis = var_4953, interleave = var_5850_interleave_0, values = (var_5778_cast_fp16, var_5780_cast_fp16, var_5782_cast_fp16, var_5784_cast_fp16))[name = string("op_5850_cast_fp16")]; bool var_5852_interleave_0 = const()[name = string("op_5852_interleave_0"), val = bool(false)]; tensor var_5852_cast_fp16 = concat(axis = var_4953, interleave = var_5852_interleave_0, values = (var_5786_cast_fp16, var_5788_cast_fp16, var_5790_cast_fp16, var_5792_cast_fp16))[name = string("op_5852_cast_fp16")]; bool var_5854_interleave_0 = const()[name = string("op_5854_interleave_0"), val = bool(false)]; tensor var_5854_cast_fp16 = concat(axis = var_4953, interleave = var_5854_interleave_0, values = (var_5794_cast_fp16, var_5796_cast_fp16, var_5798_cast_fp16, var_5800_cast_fp16))[name = string("op_5854_cast_fp16")]; bool var_5856_interleave_0 = const()[name = string("op_5856_interleave_0"), val = bool(false)]; tensor var_5856_cast_fp16 = concat(axis = var_4953, interleave = var_5856_interleave_0, values = (var_5802_cast_fp16, var_5804_cast_fp16, var_5806_cast_fp16, var_5808_cast_fp16))[name = string("op_5856_cast_fp16")]; bool var_5858_interleave_0 = const()[name = string("op_5858_interleave_0"), val = bool(false)]; tensor var_5858_cast_fp16 = concat(axis = var_4953, interleave = var_5858_interleave_0, values = (var_5810_cast_fp16, var_5812_cast_fp16, var_5814_cast_fp16, var_5816_cast_fp16))[name = string("op_5858_cast_fp16")]; bool var_5860_interleave_0 = const()[name = string("op_5860_interleave_0"), val = bool(false)]; tensor var_5860_cast_fp16 = concat(axis = var_4953, interleave = var_5860_interleave_0, values = (var_5818_cast_fp16, var_5820_cast_fp16, var_5822_cast_fp16, var_5824_cast_fp16))[name = string("op_5860_cast_fp16")]; bool var_5862_interleave_0 = const()[name = string("op_5862_interleave_0"), val = bool(false)]; tensor var_5862_cast_fp16 = concat(axis = var_4953, interleave = var_5862_interleave_0, values = (var_5826_cast_fp16, var_5828_cast_fp16, var_5830_cast_fp16, var_5832_cast_fp16))[name = string("op_5862_cast_fp16")]; bool var_5864_interleave_0 = const()[name = string("op_5864_interleave_0"), val = bool(false)]; tensor var_5864_cast_fp16 = concat(axis = var_4953, interleave = var_5864_interleave_0, values = (var_5834_cast_fp16, var_5836_cast_fp16, var_5838_cast_fp16, var_5840_cast_fp16))[name = string("op_5864_cast_fp16")]; bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; tensor input_41_cast_fp16 = concat(axis = var_4970, interleave = input_41_interleave_0, values = (var_5842_cast_fp16, var_5844_cast_fp16, var_5846_cast_fp16, var_5848_cast_fp16, var_5850_cast_fp16, var_5852_cast_fp16, var_5854_cast_fp16, var_5856_cast_fp16, var_5858_cast_fp16, var_5860_cast_fp16, var_5862_cast_fp16, var_5864_cast_fp16))[name = string("input_41_cast_fp16")]; string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80639616)))]; tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81819328)))]; tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_5883_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81820928)))]; tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81822528)))]; fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81824128)))]; tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86542784)))]; tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86548992)))]; tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91267648)))]; tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; int32 var_5912 = const()[name = string("op_5912"), val = int32(3)]; int32 var_5929 = const()[name = string("op_5929"), val = int32(1)]; tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; fp16 var_5946_to_fp16 = const()[name = string("op_5946_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_5946_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91269248)))]; tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91270848)))]; fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91272448)))]; tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92452160)))]; tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92453760)))]; tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93633472)))]; tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94813184)))]; tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; tensor var_5984_begin_0 = const()[name = string("op_5984_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5984_end_0 = const()[name = string("op_5984_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5984_end_mask_0 = const()[name = string("op_5984_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5984_cast_fp16 = slice_by_index(begin = var_5984_begin_0, end = var_5984_end_0, end_mask = var_5984_end_mask_0, x = query_13_cast_fp16)[name = string("op_5984_cast_fp16")]; tensor var_5988_begin_0 = const()[name = string("op_5988_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5988_end_0 = const()[name = string("op_5988_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5988_end_mask_0 = const()[name = string("op_5988_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5988_cast_fp16 = slice_by_index(begin = var_5988_begin_0, end = var_5988_end_0, end_mask = var_5988_end_mask_0, x = query_13_cast_fp16)[name = string("op_5988_cast_fp16")]; tensor var_5992_begin_0 = const()[name = string("op_5992_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5992_end_0 = const()[name = string("op_5992_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5992_end_mask_0 = const()[name = string("op_5992_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5992_cast_fp16 = slice_by_index(begin = var_5992_begin_0, end = var_5992_end_0, end_mask = var_5992_end_mask_0, x = query_13_cast_fp16)[name = string("op_5992_cast_fp16")]; tensor var_5996_begin_0 = const()[name = string("op_5996_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5996_end_0 = const()[name = string("op_5996_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5996_end_mask_0 = const()[name = string("op_5996_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5996_cast_fp16 = slice_by_index(begin = var_5996_begin_0, end = var_5996_end_0, end_mask = var_5996_end_mask_0, x = query_13_cast_fp16)[name = string("op_5996_cast_fp16")]; tensor var_6000_begin_0 = const()[name = string("op_6000_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6000_end_0 = const()[name = string("op_6000_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_6000_end_mask_0 = const()[name = string("op_6000_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6000_cast_fp16 = slice_by_index(begin = var_6000_begin_0, end = var_6000_end_0, end_mask = var_6000_end_mask_0, x = query_13_cast_fp16)[name = string("op_6000_cast_fp16")]; tensor var_6004_begin_0 = const()[name = string("op_6004_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6004_end_0 = const()[name = string("op_6004_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_6004_end_mask_0 = const()[name = string("op_6004_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6004_cast_fp16 = slice_by_index(begin = var_6004_begin_0, end = var_6004_end_0, end_mask = var_6004_end_mask_0, x = query_13_cast_fp16)[name = string("op_6004_cast_fp16")]; tensor var_6008_begin_0 = const()[name = string("op_6008_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6008_end_0 = const()[name = string("op_6008_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_6008_end_mask_0 = const()[name = string("op_6008_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6008_cast_fp16 = slice_by_index(begin = var_6008_begin_0, end = var_6008_end_0, end_mask = var_6008_end_mask_0, x = query_13_cast_fp16)[name = string("op_6008_cast_fp16")]; tensor var_6012_begin_0 = const()[name = string("op_6012_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6012_end_0 = const()[name = string("op_6012_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_6012_end_mask_0 = const()[name = string("op_6012_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6012_cast_fp16 = slice_by_index(begin = var_6012_begin_0, end = var_6012_end_0, end_mask = var_6012_end_mask_0, x = query_13_cast_fp16)[name = string("op_6012_cast_fp16")]; tensor var_6016_begin_0 = const()[name = string("op_6016_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6016_end_0 = const()[name = string("op_6016_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_6016_end_mask_0 = const()[name = string("op_6016_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6016_cast_fp16 = slice_by_index(begin = var_6016_begin_0, end = var_6016_end_0, end_mask = var_6016_end_mask_0, x = query_13_cast_fp16)[name = string("op_6016_cast_fp16")]; tensor var_6020_begin_0 = const()[name = string("op_6020_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6020_end_0 = const()[name = string("op_6020_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_6020_end_mask_0 = const()[name = string("op_6020_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = query_13_cast_fp16)[name = string("op_6020_cast_fp16")]; tensor var_6024_begin_0 = const()[name = string("op_6024_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6024_end_0 = const()[name = string("op_6024_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6024_end_mask_0 = const()[name = string("op_6024_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6024_cast_fp16 = slice_by_index(begin = var_6024_begin_0, end = var_6024_end_0, end_mask = var_6024_end_mask_0, x = query_13_cast_fp16)[name = string("op_6024_cast_fp16")]; tensor var_6028_begin_0 = const()[name = string("op_6028_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6028_end_0 = const()[name = string("op_6028_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6028_end_mask_0 = const()[name = string("op_6028_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6028_cast_fp16 = slice_by_index(begin = var_6028_begin_0, end = var_6028_end_0, end_mask = var_6028_end_mask_0, x = query_13_cast_fp16)[name = string("op_6028_cast_fp16")]; tensor var_6037_begin_0 = const()[name = string("op_6037_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6037_end_0 = const()[name = string("op_6037_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6037_end_mask_0 = const()[name = string("op_6037_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6037_cast_fp16 = slice_by_index(begin = var_6037_begin_0, end = var_6037_end_0, end_mask = var_6037_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6037_cast_fp16")]; tensor var_6044_begin_0 = const()[name = string("op_6044_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6044_end_0 = const()[name = string("op_6044_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6044_end_mask_0 = const()[name = string("op_6044_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6044_cast_fp16 = slice_by_index(begin = var_6044_begin_0, end = var_6044_end_0, end_mask = var_6044_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6044_cast_fp16")]; tensor var_6051_begin_0 = const()[name = string("op_6051_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6051_end_0 = const()[name = string("op_6051_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6051_end_mask_0 = const()[name = string("op_6051_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6051_cast_fp16 = slice_by_index(begin = var_6051_begin_0, end = var_6051_end_0, end_mask = var_6051_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6051_cast_fp16")]; tensor var_6058_begin_0 = const()[name = string("op_6058_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6058_end_0 = const()[name = string("op_6058_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6058_end_mask_0 = const()[name = string("op_6058_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6058_cast_fp16 = slice_by_index(begin = var_6058_begin_0, end = var_6058_end_0, end_mask = var_6058_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6058_cast_fp16")]; tensor var_6065_begin_0 = const()[name = string("op_6065_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6065_end_0 = const()[name = string("op_6065_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6065_end_mask_0 = const()[name = string("op_6065_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6065_cast_fp16 = slice_by_index(begin = var_6065_begin_0, end = var_6065_end_0, end_mask = var_6065_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6065_cast_fp16")]; tensor var_6072_begin_0 = const()[name = string("op_6072_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6072_end_0 = const()[name = string("op_6072_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6072_end_mask_0 = const()[name = string("op_6072_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6072_cast_fp16")]; tensor var_6079_begin_0 = const()[name = string("op_6079_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6079_end_0 = const()[name = string("op_6079_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6079_end_mask_0 = const()[name = string("op_6079_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6079_cast_fp16 = slice_by_index(begin = var_6079_begin_0, end = var_6079_end_0, end_mask = var_6079_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6079_cast_fp16")]; tensor var_6086_begin_0 = const()[name = string("op_6086_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6086_end_0 = const()[name = string("op_6086_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6086_end_mask_0 = const()[name = string("op_6086_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6086_cast_fp16 = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6086_cast_fp16")]; tensor var_6093_begin_0 = const()[name = string("op_6093_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6093_end_0 = const()[name = string("op_6093_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6093_end_mask_0 = const()[name = string("op_6093_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6093_cast_fp16 = slice_by_index(begin = var_6093_begin_0, end = var_6093_end_0, end_mask = var_6093_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6093_cast_fp16")]; tensor var_6100_begin_0 = const()[name = string("op_6100_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6100_end_0 = const()[name = string("op_6100_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6100_end_mask_0 = const()[name = string("op_6100_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6100_cast_fp16 = slice_by_index(begin = var_6100_begin_0, end = var_6100_end_0, end_mask = var_6100_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6100_cast_fp16")]; tensor var_6107_begin_0 = const()[name = string("op_6107_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6107_end_0 = const()[name = string("op_6107_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6107_end_mask_0 = const()[name = string("op_6107_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6107_cast_fp16 = slice_by_index(begin = var_6107_begin_0, end = var_6107_end_0, end_mask = var_6107_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6107_cast_fp16")]; tensor var_6114_begin_0 = const()[name = string("op_6114_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6114_end_0 = const()[name = string("op_6114_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6114_end_mask_0 = const()[name = string("op_6114_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6114_cast_fp16 = slice_by_index(begin = var_6114_begin_0, end = var_6114_end_0, end_mask = var_6114_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6114_cast_fp16")]; tensor var_6121_begin_0 = const()[name = string("op_6121_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6121_end_0 = const()[name = string("op_6121_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6121_end_mask_0 = const()[name = string("op_6121_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6121_cast_fp16 = slice_by_index(begin = var_6121_begin_0, end = var_6121_end_0, end_mask = var_6121_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6121_cast_fp16")]; tensor var_6128_begin_0 = const()[name = string("op_6128_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6128_end_0 = const()[name = string("op_6128_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6128_end_mask_0 = const()[name = string("op_6128_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6128_cast_fp16 = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6128_cast_fp16")]; tensor var_6135_begin_0 = const()[name = string("op_6135_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6135_end_0 = const()[name = string("op_6135_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6135_end_mask_0 = const()[name = string("op_6135_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6135_cast_fp16 = slice_by_index(begin = var_6135_begin_0, end = var_6135_end_0, end_mask = var_6135_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6135_cast_fp16")]; tensor var_6142_begin_0 = const()[name = string("op_6142_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6142_end_0 = const()[name = string("op_6142_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6142_end_mask_0 = const()[name = string("op_6142_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6142_cast_fp16 = slice_by_index(begin = var_6142_begin_0, end = var_6142_end_0, end_mask = var_6142_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6142_cast_fp16")]; tensor var_6149_begin_0 = const()[name = string("op_6149_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6149_end_0 = const()[name = string("op_6149_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6149_end_mask_0 = const()[name = string("op_6149_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6149_cast_fp16 = slice_by_index(begin = var_6149_begin_0, end = var_6149_end_0, end_mask = var_6149_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6149_cast_fp16")]; tensor var_6156_begin_0 = const()[name = string("op_6156_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6156_end_0 = const()[name = string("op_6156_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6156_end_mask_0 = const()[name = string("op_6156_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6156_cast_fp16 = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6156_cast_fp16")]; tensor var_6163_begin_0 = const()[name = string("op_6163_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6163_end_0 = const()[name = string("op_6163_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6163_end_mask_0 = const()[name = string("op_6163_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6163_cast_fp16 = slice_by_index(begin = var_6163_begin_0, end = var_6163_end_0, end_mask = var_6163_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6163_cast_fp16")]; tensor var_6170_begin_0 = const()[name = string("op_6170_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6170_end_0 = const()[name = string("op_6170_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6170_end_mask_0 = const()[name = string("op_6170_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6170_cast_fp16 = slice_by_index(begin = var_6170_begin_0, end = var_6170_end_0, end_mask = var_6170_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6170_cast_fp16")]; tensor var_6177_begin_0 = const()[name = string("op_6177_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6177_end_0 = const()[name = string("op_6177_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6177_end_mask_0 = const()[name = string("op_6177_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6177_cast_fp16 = slice_by_index(begin = var_6177_begin_0, end = var_6177_end_0, end_mask = var_6177_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6177_cast_fp16")]; tensor var_6184_begin_0 = const()[name = string("op_6184_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6184_end_0 = const()[name = string("op_6184_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6184_end_mask_0 = const()[name = string("op_6184_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6184_cast_fp16 = slice_by_index(begin = var_6184_begin_0, end = var_6184_end_0, end_mask = var_6184_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6184_cast_fp16")]; tensor var_6191_begin_0 = const()[name = string("op_6191_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6191_end_0 = const()[name = string("op_6191_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6191_end_mask_0 = const()[name = string("op_6191_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6191_cast_fp16 = slice_by_index(begin = var_6191_begin_0, end = var_6191_end_0, end_mask = var_6191_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6191_cast_fp16")]; tensor var_6198_begin_0 = const()[name = string("op_6198_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6198_end_0 = const()[name = string("op_6198_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6198_end_mask_0 = const()[name = string("op_6198_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6198_cast_fp16 = slice_by_index(begin = var_6198_begin_0, end = var_6198_end_0, end_mask = var_6198_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6198_cast_fp16")]; tensor var_6205_begin_0 = const()[name = string("op_6205_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6205_end_0 = const()[name = string("op_6205_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6205_end_mask_0 = const()[name = string("op_6205_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6205_cast_fp16 = slice_by_index(begin = var_6205_begin_0, end = var_6205_end_0, end_mask = var_6205_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6205_cast_fp16")]; tensor var_6212_begin_0 = const()[name = string("op_6212_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6212_end_0 = const()[name = string("op_6212_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6212_end_mask_0 = const()[name = string("op_6212_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6212_cast_fp16 = slice_by_index(begin = var_6212_begin_0, end = var_6212_end_0, end_mask = var_6212_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6212_cast_fp16")]; tensor var_6219_begin_0 = const()[name = string("op_6219_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6219_end_0 = const()[name = string("op_6219_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6219_end_mask_0 = const()[name = string("op_6219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6219_cast_fp16 = slice_by_index(begin = var_6219_begin_0, end = var_6219_end_0, end_mask = var_6219_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6219_cast_fp16")]; tensor var_6226_begin_0 = const()[name = string("op_6226_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6226_end_0 = const()[name = string("op_6226_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6226_end_mask_0 = const()[name = string("op_6226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6226_cast_fp16 = slice_by_index(begin = var_6226_begin_0, end = var_6226_end_0, end_mask = var_6226_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6226_cast_fp16")]; tensor var_6233_begin_0 = const()[name = string("op_6233_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6233_end_0 = const()[name = string("op_6233_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6233_end_mask_0 = const()[name = string("op_6233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6233_cast_fp16 = slice_by_index(begin = var_6233_begin_0, end = var_6233_end_0, end_mask = var_6233_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6233_cast_fp16")]; tensor var_6240_begin_0 = const()[name = string("op_6240_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6240_end_0 = const()[name = string("op_6240_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6240_end_mask_0 = const()[name = string("op_6240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6240_cast_fp16 = slice_by_index(begin = var_6240_begin_0, end = var_6240_end_0, end_mask = var_6240_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6240_cast_fp16")]; tensor var_6247_begin_0 = const()[name = string("op_6247_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6247_end_0 = const()[name = string("op_6247_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6247_end_mask_0 = const()[name = string("op_6247_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6247_cast_fp16 = slice_by_index(begin = var_6247_begin_0, end = var_6247_end_0, end_mask = var_6247_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6247_cast_fp16")]; tensor var_6254_begin_0 = const()[name = string("op_6254_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6254_end_0 = const()[name = string("op_6254_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6254_end_mask_0 = const()[name = string("op_6254_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6254_cast_fp16 = slice_by_index(begin = var_6254_begin_0, end = var_6254_end_0, end_mask = var_6254_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6254_cast_fp16")]; tensor var_6261_begin_0 = const()[name = string("op_6261_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6261_end_0 = const()[name = string("op_6261_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6261_end_mask_0 = const()[name = string("op_6261_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6261_cast_fp16 = slice_by_index(begin = var_6261_begin_0, end = var_6261_end_0, end_mask = var_6261_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6261_cast_fp16")]; tensor var_6268_begin_0 = const()[name = string("op_6268_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6268_end_0 = const()[name = string("op_6268_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6268_end_mask_0 = const()[name = string("op_6268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6268_cast_fp16 = slice_by_index(begin = var_6268_begin_0, end = var_6268_end_0, end_mask = var_6268_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6268_cast_fp16")]; tensor var_6275_begin_0 = const()[name = string("op_6275_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6275_end_0 = const()[name = string("op_6275_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6275_end_mask_0 = const()[name = string("op_6275_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6275_cast_fp16 = slice_by_index(begin = var_6275_begin_0, end = var_6275_end_0, end_mask = var_6275_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6275_cast_fp16")]; tensor var_6282_begin_0 = const()[name = string("op_6282_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6282_end_0 = const()[name = string("op_6282_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6282_end_mask_0 = const()[name = string("op_6282_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6282_cast_fp16 = slice_by_index(begin = var_6282_begin_0, end = var_6282_end_0, end_mask = var_6282_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6282_cast_fp16")]; tensor var_6289_begin_0 = const()[name = string("op_6289_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6289_end_0 = const()[name = string("op_6289_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6289_end_mask_0 = const()[name = string("op_6289_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6289_cast_fp16 = slice_by_index(begin = var_6289_begin_0, end = var_6289_end_0, end_mask = var_6289_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6289_cast_fp16")]; tensor var_6296_begin_0 = const()[name = string("op_6296_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6296_end_0 = const()[name = string("op_6296_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6296_end_mask_0 = const()[name = string("op_6296_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6296_cast_fp16 = slice_by_index(begin = var_6296_begin_0, end = var_6296_end_0, end_mask = var_6296_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6296_cast_fp16")]; tensor var_6303_begin_0 = const()[name = string("op_6303_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6303_end_0 = const()[name = string("op_6303_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6303_end_mask_0 = const()[name = string("op_6303_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6303_cast_fp16 = slice_by_index(begin = var_6303_begin_0, end = var_6303_end_0, end_mask = var_6303_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6303_cast_fp16")]; tensor var_6310_begin_0 = const()[name = string("op_6310_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6310_end_0 = const()[name = string("op_6310_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6310_end_mask_0 = const()[name = string("op_6310_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6310_cast_fp16 = slice_by_index(begin = var_6310_begin_0, end = var_6310_end_0, end_mask = var_6310_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6310_cast_fp16")]; tensor var_6317_begin_0 = const()[name = string("op_6317_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6317_end_0 = const()[name = string("op_6317_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6317_end_mask_0 = const()[name = string("op_6317_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6317_cast_fp16 = slice_by_index(begin = var_6317_begin_0, end = var_6317_end_0, end_mask = var_6317_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6317_cast_fp16")]; tensor var_6324_begin_0 = const()[name = string("op_6324_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6324_end_0 = const()[name = string("op_6324_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6324_end_mask_0 = const()[name = string("op_6324_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6324_cast_fp16 = slice_by_index(begin = var_6324_begin_0, end = var_6324_end_0, end_mask = var_6324_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6324_cast_fp16")]; tensor var_6331_begin_0 = const()[name = string("op_6331_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6331_end_0 = const()[name = string("op_6331_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6331_end_mask_0 = const()[name = string("op_6331_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6331_cast_fp16 = slice_by_index(begin = var_6331_begin_0, end = var_6331_end_0, end_mask = var_6331_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6331_cast_fp16")]; tensor var_6338_begin_0 = const()[name = string("op_6338_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6338_end_0 = const()[name = string("op_6338_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6338_end_mask_0 = const()[name = string("op_6338_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6338_cast_fp16 = slice_by_index(begin = var_6338_begin_0, end = var_6338_end_0, end_mask = var_6338_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6338_cast_fp16")]; tensor var_6345_begin_0 = const()[name = string("op_6345_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6345_end_0 = const()[name = string("op_6345_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6345_end_mask_0 = const()[name = string("op_6345_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6345_cast_fp16 = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6345_cast_fp16")]; tensor var_6352_begin_0 = const()[name = string("op_6352_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6352_end_0 = const()[name = string("op_6352_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6352_end_mask_0 = const()[name = string("op_6352_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6352_cast_fp16 = slice_by_index(begin = var_6352_begin_0, end = var_6352_end_0, end_mask = var_6352_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6352_cast_fp16")]; tensor var_6359_begin_0 = const()[name = string("op_6359_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6359_end_0 = const()[name = string("op_6359_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6359_end_mask_0 = const()[name = string("op_6359_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6359_cast_fp16 = slice_by_index(begin = var_6359_begin_0, end = var_6359_end_0, end_mask = var_6359_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6359_cast_fp16")]; tensor var_6366_begin_0 = const()[name = string("op_6366_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6366_end_0 = const()[name = string("op_6366_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6366_end_mask_0 = const()[name = string("op_6366_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6366_cast_fp16 = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6366_cast_fp16")]; tensor k_13_perm_0 = const()[name = string("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6371_begin_0 = const()[name = string("op_6371_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6371_end_0 = const()[name = string("op_6371_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_6371_end_mask_0 = const()[name = string("op_6371_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_13_cast_fp16 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = string("transpose_5")]; tensor var_6371_cast_fp16 = slice_by_index(begin = var_6371_begin_0, end = var_6371_end_0, end_mask = var_6371_end_mask_0, x = k_13_cast_fp16)[name = string("op_6371_cast_fp16")]; tensor var_6375_begin_0 = const()[name = string("op_6375_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_6375_end_0 = const()[name = string("op_6375_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_6375_end_mask_0 = const()[name = string("op_6375_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6375_cast_fp16 = slice_by_index(begin = var_6375_begin_0, end = var_6375_end_0, end_mask = var_6375_end_mask_0, x = k_13_cast_fp16)[name = string("op_6375_cast_fp16")]; tensor var_6379_begin_0 = const()[name = string("op_6379_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_6379_end_0 = const()[name = string("op_6379_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_6379_end_mask_0 = const()[name = string("op_6379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6379_cast_fp16 = slice_by_index(begin = var_6379_begin_0, end = var_6379_end_0, end_mask = var_6379_end_mask_0, x = k_13_cast_fp16)[name = string("op_6379_cast_fp16")]; tensor var_6383_begin_0 = const()[name = string("op_6383_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_6383_end_0 = const()[name = string("op_6383_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_6383_end_mask_0 = const()[name = string("op_6383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6383_cast_fp16 = slice_by_index(begin = var_6383_begin_0, end = var_6383_end_0, end_mask = var_6383_end_mask_0, x = k_13_cast_fp16)[name = string("op_6383_cast_fp16")]; tensor var_6387_begin_0 = const()[name = string("op_6387_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_6387_end_0 = const()[name = string("op_6387_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_6387_end_mask_0 = const()[name = string("op_6387_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6387_cast_fp16 = slice_by_index(begin = var_6387_begin_0, end = var_6387_end_0, end_mask = var_6387_end_mask_0, x = k_13_cast_fp16)[name = string("op_6387_cast_fp16")]; tensor var_6391_begin_0 = const()[name = string("op_6391_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6391_end_0 = const()[name = string("op_6391_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_6391_end_mask_0 = const()[name = string("op_6391_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6391_cast_fp16 = slice_by_index(begin = var_6391_begin_0, end = var_6391_end_0, end_mask = var_6391_end_mask_0, x = k_13_cast_fp16)[name = string("op_6391_cast_fp16")]; tensor var_6395_begin_0 = const()[name = string("op_6395_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_6395_end_0 = const()[name = string("op_6395_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_6395_end_mask_0 = const()[name = string("op_6395_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6395_cast_fp16 = slice_by_index(begin = var_6395_begin_0, end = var_6395_end_0, end_mask = var_6395_end_mask_0, x = k_13_cast_fp16)[name = string("op_6395_cast_fp16")]; tensor var_6399_begin_0 = const()[name = string("op_6399_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_6399_end_0 = const()[name = string("op_6399_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_6399_end_mask_0 = const()[name = string("op_6399_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6399_cast_fp16 = slice_by_index(begin = var_6399_begin_0, end = var_6399_end_0, end_mask = var_6399_end_mask_0, x = k_13_cast_fp16)[name = string("op_6399_cast_fp16")]; tensor var_6403_begin_0 = const()[name = string("op_6403_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6403_end_0 = const()[name = string("op_6403_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_6403_end_mask_0 = const()[name = string("op_6403_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6403_cast_fp16 = slice_by_index(begin = var_6403_begin_0, end = var_6403_end_0, end_mask = var_6403_end_mask_0, x = k_13_cast_fp16)[name = string("op_6403_cast_fp16")]; tensor var_6407_begin_0 = const()[name = string("op_6407_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_6407_end_0 = const()[name = string("op_6407_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_6407_end_mask_0 = const()[name = string("op_6407_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6407_cast_fp16 = slice_by_index(begin = var_6407_begin_0, end = var_6407_end_0, end_mask = var_6407_end_mask_0, x = k_13_cast_fp16)[name = string("op_6407_cast_fp16")]; tensor var_6411_begin_0 = const()[name = string("op_6411_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6411_end_0 = const()[name = string("op_6411_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_6411_end_mask_0 = const()[name = string("op_6411_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6411_cast_fp16 = slice_by_index(begin = var_6411_begin_0, end = var_6411_end_0, end_mask = var_6411_end_mask_0, x = k_13_cast_fp16)[name = string("op_6411_cast_fp16")]; tensor var_6415_begin_0 = const()[name = string("op_6415_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_6415_end_0 = const()[name = string("op_6415_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_6415_end_mask_0 = const()[name = string("op_6415_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6415_cast_fp16 = slice_by_index(begin = var_6415_begin_0, end = var_6415_end_0, end_mask = var_6415_end_mask_0, x = k_13_cast_fp16)[name = string("op_6415_cast_fp16")]; tensor var_6417_begin_0 = const()[name = string("op_6417_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6417_end_0 = const()[name = string("op_6417_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6417_end_mask_0 = const()[name = string("op_6417_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6417_cast_fp16 = slice_by_index(begin = var_6417_begin_0, end = var_6417_end_0, end_mask = var_6417_end_mask_0, x = value_13_cast_fp16)[name = string("op_6417_cast_fp16")]; tensor var_6421_begin_0 = const()[name = string("op_6421_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6421_end_0 = const()[name = string("op_6421_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_6421_end_mask_0 = const()[name = string("op_6421_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6421_cast_fp16 = slice_by_index(begin = var_6421_begin_0, end = var_6421_end_0, end_mask = var_6421_end_mask_0, x = value_13_cast_fp16)[name = string("op_6421_cast_fp16")]; tensor var_6425_begin_0 = const()[name = string("op_6425_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6425_end_0 = const()[name = string("op_6425_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_6425_end_mask_0 = const()[name = string("op_6425_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6425_cast_fp16 = slice_by_index(begin = var_6425_begin_0, end = var_6425_end_0, end_mask = var_6425_end_mask_0, x = value_13_cast_fp16)[name = string("op_6425_cast_fp16")]; tensor var_6429_begin_0 = const()[name = string("op_6429_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6429_end_0 = const()[name = string("op_6429_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_6429_end_mask_0 = const()[name = string("op_6429_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6429_cast_fp16 = slice_by_index(begin = var_6429_begin_0, end = var_6429_end_0, end_mask = var_6429_end_mask_0, x = value_13_cast_fp16)[name = string("op_6429_cast_fp16")]; tensor var_6433_begin_0 = const()[name = string("op_6433_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6433_end_0 = const()[name = string("op_6433_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_6433_end_mask_0 = const()[name = string("op_6433_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6433_cast_fp16 = slice_by_index(begin = var_6433_begin_0, end = var_6433_end_0, end_mask = var_6433_end_mask_0, x = value_13_cast_fp16)[name = string("op_6433_cast_fp16")]; tensor var_6437_begin_0 = const()[name = string("op_6437_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6437_end_0 = const()[name = string("op_6437_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_6437_end_mask_0 = const()[name = string("op_6437_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6437_cast_fp16 = slice_by_index(begin = var_6437_begin_0, end = var_6437_end_0, end_mask = var_6437_end_mask_0, x = value_13_cast_fp16)[name = string("op_6437_cast_fp16")]; tensor var_6441_begin_0 = const()[name = string("op_6441_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6441_end_0 = const()[name = string("op_6441_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_6441_end_mask_0 = const()[name = string("op_6441_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6441_cast_fp16 = slice_by_index(begin = var_6441_begin_0, end = var_6441_end_0, end_mask = var_6441_end_mask_0, x = value_13_cast_fp16)[name = string("op_6441_cast_fp16")]; tensor var_6445_begin_0 = const()[name = string("op_6445_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6445_end_0 = const()[name = string("op_6445_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_6445_end_mask_0 = const()[name = string("op_6445_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6445_cast_fp16 = slice_by_index(begin = var_6445_begin_0, end = var_6445_end_0, end_mask = var_6445_end_mask_0, x = value_13_cast_fp16)[name = string("op_6445_cast_fp16")]; tensor var_6449_begin_0 = const()[name = string("op_6449_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6449_end_0 = const()[name = string("op_6449_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_6449_end_mask_0 = const()[name = string("op_6449_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6449_cast_fp16 = slice_by_index(begin = var_6449_begin_0, end = var_6449_end_0, end_mask = var_6449_end_mask_0, x = value_13_cast_fp16)[name = string("op_6449_cast_fp16")]; tensor var_6453_begin_0 = const()[name = string("op_6453_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6453_end_0 = const()[name = string("op_6453_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_6453_end_mask_0 = const()[name = string("op_6453_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6453_cast_fp16 = slice_by_index(begin = var_6453_begin_0, end = var_6453_end_0, end_mask = var_6453_end_mask_0, x = value_13_cast_fp16)[name = string("op_6453_cast_fp16")]; tensor var_6457_begin_0 = const()[name = string("op_6457_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6457_end_0 = const()[name = string("op_6457_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6457_end_mask_0 = const()[name = string("op_6457_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6457_cast_fp16 = slice_by_index(begin = var_6457_begin_0, end = var_6457_end_0, end_mask = var_6457_end_mask_0, x = value_13_cast_fp16)[name = string("op_6457_cast_fp16")]; tensor var_6461_begin_0 = const()[name = string("op_6461_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6461_end_0 = const()[name = string("op_6461_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6461_end_mask_0 = const()[name = string("op_6461_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6461_cast_fp16 = slice_by_index(begin = var_6461_begin_0, end = var_6461_end_0, end_mask = var_6461_end_mask_0, x = value_13_cast_fp16)[name = string("op_6461_cast_fp16")]; string _SplitHeadsQ__mh_w_577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_577_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_577_equation_0, values = (var_6371_cast_fp16, var_6037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_577_cast_fp16")]; string _SplitHeadsQ__mh_w_579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_579_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_579_equation_0, values = (var_6371_cast_fp16, var_6044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_579_cast_fp16")]; string _SplitHeadsQ__mh_w_581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_581_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_581_equation_0, values = (var_6371_cast_fp16, var_6051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_581_cast_fp16")]; string _SplitHeadsQ__mh_w_583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_583_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_583_equation_0, values = (var_6371_cast_fp16, var_6058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_583_cast_fp16")]; string _SplitHeadsQ__mh_w_585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_585_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_585_equation_0, values = (var_6375_cast_fp16, var_6065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_585_cast_fp16")]; string _SplitHeadsQ__mh_w_587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_587_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_587_equation_0, values = (var_6375_cast_fp16, var_6072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_587_cast_fp16")]; string _SplitHeadsQ__mh_w_589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_589_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_589_equation_0, values = (var_6375_cast_fp16, var_6079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_589_cast_fp16")]; string _SplitHeadsQ__mh_w_591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_591_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_591_equation_0, values = (var_6375_cast_fp16, var_6086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_591_cast_fp16")]; string _SplitHeadsQ__mh_w_593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_593_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_593_equation_0, values = (var_6379_cast_fp16, var_6093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_593_cast_fp16")]; string _SplitHeadsQ__mh_w_595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_595_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_595_equation_0, values = (var_6379_cast_fp16, var_6100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_595_cast_fp16")]; string _SplitHeadsQ__mh_w_597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_597_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_597_equation_0, values = (var_6379_cast_fp16, var_6107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_597_cast_fp16")]; string _SplitHeadsQ__mh_w_599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_599_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_599_equation_0, values = (var_6379_cast_fp16, var_6114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_599_cast_fp16")]; string _SplitHeadsQ__mh_w_601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_601_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_601_equation_0, values = (var_6383_cast_fp16, var_6121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_601_cast_fp16")]; string _SplitHeadsQ__mh_w_603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_603_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_603_equation_0, values = (var_6383_cast_fp16, var_6128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_603_cast_fp16")]; string _SplitHeadsQ__mh_w_605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_605_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_605_equation_0, values = (var_6383_cast_fp16, var_6135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_605_cast_fp16")]; string _SplitHeadsQ__mh_w_607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_607_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_607_equation_0, values = (var_6383_cast_fp16, var_6142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_607_cast_fp16")]; string _SplitHeadsQ__mh_w_609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_609_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_609_equation_0, values = (var_6387_cast_fp16, var_6149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_609_cast_fp16")]; string _SplitHeadsQ__mh_w_611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_611_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_611_equation_0, values = (var_6387_cast_fp16, var_6156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_611_cast_fp16")]; string _SplitHeadsQ__mh_w_613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_613_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_613_equation_0, values = (var_6387_cast_fp16, var_6163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_613_cast_fp16")]; string _SplitHeadsQ__mh_w_615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_615_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_615_equation_0, values = (var_6387_cast_fp16, var_6170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_615_cast_fp16")]; string _SplitHeadsQ__mh_w_617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_617_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_617_equation_0, values = (var_6391_cast_fp16, var_6177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_617_cast_fp16")]; string _SplitHeadsQ__mh_w_619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_619_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_619_equation_0, values = (var_6391_cast_fp16, var_6184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_619_cast_fp16")]; string _SplitHeadsQ__mh_w_621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_621_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_621_equation_0, values = (var_6391_cast_fp16, var_6191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_621_cast_fp16")]; string _SplitHeadsQ__mh_w_623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_623_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_623_equation_0, values = (var_6391_cast_fp16, var_6198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_623_cast_fp16")]; string _SplitHeadsQ__mh_w_625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_625_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_625_equation_0, values = (var_6395_cast_fp16, var_6205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_625_cast_fp16")]; string _SplitHeadsQ__mh_w_627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_627_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_627_equation_0, values = (var_6395_cast_fp16, var_6212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_627_cast_fp16")]; string _SplitHeadsQ__mh_w_629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_629_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_629_equation_0, values = (var_6395_cast_fp16, var_6219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_629_cast_fp16")]; string _SplitHeadsQ__mh_w_631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_631_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_631_equation_0, values = (var_6395_cast_fp16, var_6226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_631_cast_fp16")]; string _SplitHeadsQ__mh_w_633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_633_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_633_equation_0, values = (var_6399_cast_fp16, var_6233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_633_cast_fp16")]; string _SplitHeadsQ__mh_w_635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_635_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_635_equation_0, values = (var_6399_cast_fp16, var_6240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_635_cast_fp16")]; string _SplitHeadsQ__mh_w_637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_637_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_637_equation_0, values = (var_6399_cast_fp16, var_6247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_637_cast_fp16")]; string _SplitHeadsQ__mh_w_639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_639_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_639_equation_0, values = (var_6399_cast_fp16, var_6254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_639_cast_fp16")]; string _SplitHeadsQ__mh_w_641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_641_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_641_equation_0, values = (var_6403_cast_fp16, var_6261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_641_cast_fp16")]; string _SplitHeadsQ__mh_w_643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_643_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_643_equation_0, values = (var_6403_cast_fp16, var_6268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_643_cast_fp16")]; string _SplitHeadsQ__mh_w_645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_645_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_645_equation_0, values = (var_6403_cast_fp16, var_6275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_645_cast_fp16")]; string _SplitHeadsQ__mh_w_647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_647_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_647_equation_0, values = (var_6403_cast_fp16, var_6282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_647_cast_fp16")]; string _SplitHeadsQ__mh_w_649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_649_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_649_equation_0, values = (var_6407_cast_fp16, var_6289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_649_cast_fp16")]; string _SplitHeadsQ__mh_w_651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_651_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_651_equation_0, values = (var_6407_cast_fp16, var_6296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_651_cast_fp16")]; string _SplitHeadsQ__mh_w_653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_653_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_653_equation_0, values = (var_6407_cast_fp16, var_6303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_653_cast_fp16")]; string _SplitHeadsQ__mh_w_655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_655_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_655_equation_0, values = (var_6407_cast_fp16, var_6310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_655_cast_fp16")]; string _SplitHeadsQ__mh_w_657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_657_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_657_equation_0, values = (var_6411_cast_fp16, var_6317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_657_cast_fp16")]; string _SplitHeadsQ__mh_w_659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_659_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_659_equation_0, values = (var_6411_cast_fp16, var_6324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_659_cast_fp16")]; string _SplitHeadsQ__mh_w_661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_661_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_661_equation_0, values = (var_6411_cast_fp16, var_6331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_661_cast_fp16")]; string _SplitHeadsQ__mh_w_663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_663_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_663_equation_0, values = (var_6411_cast_fp16, var_6338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_663_cast_fp16")]; string _SplitHeadsQ__mh_w_665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_665_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_665_equation_0, values = (var_6415_cast_fp16, var_6345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_665_cast_fp16")]; string _SplitHeadsQ__mh_w_667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_667_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_667_equation_0, values = (var_6415_cast_fp16, var_6352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_667_cast_fp16")]; string _SplitHeadsQ__mh_w_669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_669_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_669_equation_0, values = (var_6415_cast_fp16, var_6359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_669_cast_fp16")]; string _SplitHeadsQ__mh_w_671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_671_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_671_equation_0, values = (var_6415_cast_fp16, var_6366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_671_cast_fp16")]; fp16 var_6560_to_fp16 = const()[name = string("op_6560_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_577_cast_fp16, y = var_6560_to_fp16)[name = string("aw_chunk_577_cast_fp16")]; fp16 var_6562_to_fp16 = const()[name = string("op_6562_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_579_cast_fp16, y = var_6562_to_fp16)[name = string("aw_chunk_579_cast_fp16")]; fp16 var_6564_to_fp16 = const()[name = string("op_6564_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_581_cast_fp16, y = var_6564_to_fp16)[name = string("aw_chunk_581_cast_fp16")]; fp16 var_6566_to_fp16 = const()[name = string("op_6566_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_583_cast_fp16, y = var_6566_to_fp16)[name = string("aw_chunk_583_cast_fp16")]; fp16 var_6568_to_fp16 = const()[name = string("op_6568_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_585_cast_fp16, y = var_6568_to_fp16)[name = string("aw_chunk_585_cast_fp16")]; fp16 var_6570_to_fp16 = const()[name = string("op_6570_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_587_cast_fp16, y = var_6570_to_fp16)[name = string("aw_chunk_587_cast_fp16")]; fp16 var_6572_to_fp16 = const()[name = string("op_6572_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_589_cast_fp16, y = var_6572_to_fp16)[name = string("aw_chunk_589_cast_fp16")]; fp16 var_6574_to_fp16 = const()[name = string("op_6574_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_591_cast_fp16, y = var_6574_to_fp16)[name = string("aw_chunk_591_cast_fp16")]; fp16 var_6576_to_fp16 = const()[name = string("op_6576_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_593_cast_fp16, y = var_6576_to_fp16)[name = string("aw_chunk_593_cast_fp16")]; fp16 var_6578_to_fp16 = const()[name = string("op_6578_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_595_cast_fp16, y = var_6578_to_fp16)[name = string("aw_chunk_595_cast_fp16")]; fp16 var_6580_to_fp16 = const()[name = string("op_6580_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_597_cast_fp16, y = var_6580_to_fp16)[name = string("aw_chunk_597_cast_fp16")]; fp16 var_6582_to_fp16 = const()[name = string("op_6582_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_599_cast_fp16, y = var_6582_to_fp16)[name = string("aw_chunk_599_cast_fp16")]; fp16 var_6584_to_fp16 = const()[name = string("op_6584_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_601_cast_fp16, y = var_6584_to_fp16)[name = string("aw_chunk_601_cast_fp16")]; fp16 var_6586_to_fp16 = const()[name = string("op_6586_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_603_cast_fp16, y = var_6586_to_fp16)[name = string("aw_chunk_603_cast_fp16")]; fp16 var_6588_to_fp16 = const()[name = string("op_6588_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_605_cast_fp16, y = var_6588_to_fp16)[name = string("aw_chunk_605_cast_fp16")]; fp16 var_6590_to_fp16 = const()[name = string("op_6590_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_607_cast_fp16, y = var_6590_to_fp16)[name = string("aw_chunk_607_cast_fp16")]; fp16 var_6592_to_fp16 = const()[name = string("op_6592_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_609_cast_fp16, y = var_6592_to_fp16)[name = string("aw_chunk_609_cast_fp16")]; fp16 var_6594_to_fp16 = const()[name = string("op_6594_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_611_cast_fp16, y = var_6594_to_fp16)[name = string("aw_chunk_611_cast_fp16")]; fp16 var_6596_to_fp16 = const()[name = string("op_6596_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_613_cast_fp16, y = var_6596_to_fp16)[name = string("aw_chunk_613_cast_fp16")]; fp16 var_6598_to_fp16 = const()[name = string("op_6598_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_615_cast_fp16, y = var_6598_to_fp16)[name = string("aw_chunk_615_cast_fp16")]; fp16 var_6600_to_fp16 = const()[name = string("op_6600_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_617_cast_fp16, y = var_6600_to_fp16)[name = string("aw_chunk_617_cast_fp16")]; fp16 var_6602_to_fp16 = const()[name = string("op_6602_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_619_cast_fp16, y = var_6602_to_fp16)[name = string("aw_chunk_619_cast_fp16")]; fp16 var_6604_to_fp16 = const()[name = string("op_6604_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_621_cast_fp16, y = var_6604_to_fp16)[name = string("aw_chunk_621_cast_fp16")]; fp16 var_6606_to_fp16 = const()[name = string("op_6606_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_623_cast_fp16, y = var_6606_to_fp16)[name = string("aw_chunk_623_cast_fp16")]; fp16 var_6608_to_fp16 = const()[name = string("op_6608_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_625_cast_fp16, y = var_6608_to_fp16)[name = string("aw_chunk_625_cast_fp16")]; fp16 var_6610_to_fp16 = const()[name = string("op_6610_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_627_cast_fp16, y = var_6610_to_fp16)[name = string("aw_chunk_627_cast_fp16")]; fp16 var_6612_to_fp16 = const()[name = string("op_6612_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_629_cast_fp16, y = var_6612_to_fp16)[name = string("aw_chunk_629_cast_fp16")]; fp16 var_6614_to_fp16 = const()[name = string("op_6614_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_631_cast_fp16, y = var_6614_to_fp16)[name = string("aw_chunk_631_cast_fp16")]; fp16 var_6616_to_fp16 = const()[name = string("op_6616_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_633_cast_fp16, y = var_6616_to_fp16)[name = string("aw_chunk_633_cast_fp16")]; fp16 var_6618_to_fp16 = const()[name = string("op_6618_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_635_cast_fp16, y = var_6618_to_fp16)[name = string("aw_chunk_635_cast_fp16")]; fp16 var_6620_to_fp16 = const()[name = string("op_6620_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_637_cast_fp16, y = var_6620_to_fp16)[name = string("aw_chunk_637_cast_fp16")]; fp16 var_6622_to_fp16 = const()[name = string("op_6622_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_639_cast_fp16, y = var_6622_to_fp16)[name = string("aw_chunk_639_cast_fp16")]; fp16 var_6624_to_fp16 = const()[name = string("op_6624_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_641_cast_fp16, y = var_6624_to_fp16)[name = string("aw_chunk_641_cast_fp16")]; fp16 var_6626_to_fp16 = const()[name = string("op_6626_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_643_cast_fp16, y = var_6626_to_fp16)[name = string("aw_chunk_643_cast_fp16")]; fp16 var_6628_to_fp16 = const()[name = string("op_6628_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_645_cast_fp16, y = var_6628_to_fp16)[name = string("aw_chunk_645_cast_fp16")]; fp16 var_6630_to_fp16 = const()[name = string("op_6630_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_647_cast_fp16, y = var_6630_to_fp16)[name = string("aw_chunk_647_cast_fp16")]; fp16 var_6632_to_fp16 = const()[name = string("op_6632_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_649_cast_fp16, y = var_6632_to_fp16)[name = string("aw_chunk_649_cast_fp16")]; fp16 var_6634_to_fp16 = const()[name = string("op_6634_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_651_cast_fp16, y = var_6634_to_fp16)[name = string("aw_chunk_651_cast_fp16")]; fp16 var_6636_to_fp16 = const()[name = string("op_6636_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_653_cast_fp16, y = var_6636_to_fp16)[name = string("aw_chunk_653_cast_fp16")]; fp16 var_6638_to_fp16 = const()[name = string("op_6638_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_655_cast_fp16, y = var_6638_to_fp16)[name = string("aw_chunk_655_cast_fp16")]; fp16 var_6640_to_fp16 = const()[name = string("op_6640_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_657_cast_fp16, y = var_6640_to_fp16)[name = string("aw_chunk_657_cast_fp16")]; fp16 var_6642_to_fp16 = const()[name = string("op_6642_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_659_cast_fp16, y = var_6642_to_fp16)[name = string("aw_chunk_659_cast_fp16")]; fp16 var_6644_to_fp16 = const()[name = string("op_6644_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_661_cast_fp16, y = var_6644_to_fp16)[name = string("aw_chunk_661_cast_fp16")]; fp16 var_6646_to_fp16 = const()[name = string("op_6646_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_663_cast_fp16, y = var_6646_to_fp16)[name = string("aw_chunk_663_cast_fp16")]; fp16 var_6648_to_fp16 = const()[name = string("op_6648_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_665_cast_fp16, y = var_6648_to_fp16)[name = string("aw_chunk_665_cast_fp16")]; fp16 var_6650_to_fp16 = const()[name = string("op_6650_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_667_cast_fp16, y = var_6650_to_fp16)[name = string("aw_chunk_667_cast_fp16")]; fp16 var_6652_to_fp16 = const()[name = string("op_6652_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_669_cast_fp16, y = var_6652_to_fp16)[name = string("aw_chunk_669_cast_fp16")]; fp16 var_6654_to_fp16 = const()[name = string("op_6654_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_671_cast_fp16, y = var_6654_to_fp16)[name = string("aw_chunk_671_cast_fp16")]; tensor var_6656_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_577_cast_fp16)[name = string("op_6656_cast_fp16")]; tensor var_6657_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_579_cast_fp16)[name = string("op_6657_cast_fp16")]; tensor var_6658_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_581_cast_fp16)[name = string("op_6658_cast_fp16")]; tensor var_6659_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_583_cast_fp16)[name = string("op_6659_cast_fp16")]; tensor var_6660_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_585_cast_fp16)[name = string("op_6660_cast_fp16")]; tensor var_6661_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_587_cast_fp16)[name = string("op_6661_cast_fp16")]; tensor var_6662_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_589_cast_fp16)[name = string("op_6662_cast_fp16")]; tensor var_6663_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_591_cast_fp16)[name = string("op_6663_cast_fp16")]; tensor var_6664_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_593_cast_fp16)[name = string("op_6664_cast_fp16")]; tensor var_6665_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_595_cast_fp16)[name = string("op_6665_cast_fp16")]; tensor var_6666_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_597_cast_fp16)[name = string("op_6666_cast_fp16")]; tensor var_6667_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_599_cast_fp16)[name = string("op_6667_cast_fp16")]; tensor var_6668_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_601_cast_fp16)[name = string("op_6668_cast_fp16")]; tensor var_6669_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_603_cast_fp16)[name = string("op_6669_cast_fp16")]; tensor var_6670_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_605_cast_fp16)[name = string("op_6670_cast_fp16")]; tensor var_6671_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_607_cast_fp16)[name = string("op_6671_cast_fp16")]; tensor var_6672_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_609_cast_fp16)[name = string("op_6672_cast_fp16")]; tensor var_6673_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_611_cast_fp16)[name = string("op_6673_cast_fp16")]; tensor var_6674_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_613_cast_fp16)[name = string("op_6674_cast_fp16")]; tensor var_6675_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_615_cast_fp16)[name = string("op_6675_cast_fp16")]; tensor var_6676_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_617_cast_fp16)[name = string("op_6676_cast_fp16")]; tensor var_6677_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_619_cast_fp16)[name = string("op_6677_cast_fp16")]; tensor var_6678_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_621_cast_fp16)[name = string("op_6678_cast_fp16")]; tensor var_6679_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_623_cast_fp16)[name = string("op_6679_cast_fp16")]; tensor var_6680_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_625_cast_fp16)[name = string("op_6680_cast_fp16")]; tensor var_6681_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_627_cast_fp16)[name = string("op_6681_cast_fp16")]; tensor var_6682_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_629_cast_fp16)[name = string("op_6682_cast_fp16")]; tensor var_6683_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_631_cast_fp16)[name = string("op_6683_cast_fp16")]; tensor var_6684_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_633_cast_fp16)[name = string("op_6684_cast_fp16")]; tensor var_6685_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_635_cast_fp16)[name = string("op_6685_cast_fp16")]; tensor var_6686_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_637_cast_fp16)[name = string("op_6686_cast_fp16")]; tensor var_6687_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_639_cast_fp16)[name = string("op_6687_cast_fp16")]; tensor var_6688_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_641_cast_fp16)[name = string("op_6688_cast_fp16")]; tensor var_6689_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_643_cast_fp16)[name = string("op_6689_cast_fp16")]; tensor var_6690_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_645_cast_fp16)[name = string("op_6690_cast_fp16")]; tensor var_6691_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_647_cast_fp16)[name = string("op_6691_cast_fp16")]; tensor var_6692_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_649_cast_fp16)[name = string("op_6692_cast_fp16")]; tensor var_6693_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_651_cast_fp16)[name = string("op_6693_cast_fp16")]; tensor var_6694_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_653_cast_fp16)[name = string("op_6694_cast_fp16")]; tensor var_6695_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_655_cast_fp16)[name = string("op_6695_cast_fp16")]; tensor var_6696_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_657_cast_fp16)[name = string("op_6696_cast_fp16")]; tensor var_6697_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_659_cast_fp16)[name = string("op_6697_cast_fp16")]; tensor var_6698_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_661_cast_fp16)[name = string("op_6698_cast_fp16")]; tensor var_6699_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_663_cast_fp16)[name = string("op_6699_cast_fp16")]; tensor var_6700_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_665_cast_fp16)[name = string("op_6700_cast_fp16")]; tensor var_6701_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_667_cast_fp16)[name = string("op_6701_cast_fp16")]; tensor var_6702_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_669_cast_fp16)[name = string("op_6702_cast_fp16")]; tensor var_6703_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_671_cast_fp16)[name = string("op_6703_cast_fp16")]; string var_6705_equation_0 = const()[name = string("op_6705_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6417_cast_fp16, var_6656_cast_fp16))[name = string("op_6705_cast_fp16")]; string var_6707_equation_0 = const()[name = string("op_6707_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6707_cast_fp16 = einsum(equation = var_6707_equation_0, values = (var_6417_cast_fp16, var_6657_cast_fp16))[name = string("op_6707_cast_fp16")]; string var_6709_equation_0 = const()[name = string("op_6709_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6417_cast_fp16, var_6658_cast_fp16))[name = string("op_6709_cast_fp16")]; string var_6711_equation_0 = const()[name = string("op_6711_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6711_cast_fp16 = einsum(equation = var_6711_equation_0, values = (var_6417_cast_fp16, var_6659_cast_fp16))[name = string("op_6711_cast_fp16")]; string var_6713_equation_0 = const()[name = string("op_6713_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6421_cast_fp16, var_6660_cast_fp16))[name = string("op_6713_cast_fp16")]; string var_6715_equation_0 = const()[name = string("op_6715_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6715_cast_fp16 = einsum(equation = var_6715_equation_0, values = (var_6421_cast_fp16, var_6661_cast_fp16))[name = string("op_6715_cast_fp16")]; string var_6717_equation_0 = const()[name = string("op_6717_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6421_cast_fp16, var_6662_cast_fp16))[name = string("op_6717_cast_fp16")]; string var_6719_equation_0 = const()[name = string("op_6719_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6719_cast_fp16 = einsum(equation = var_6719_equation_0, values = (var_6421_cast_fp16, var_6663_cast_fp16))[name = string("op_6719_cast_fp16")]; string var_6721_equation_0 = const()[name = string("op_6721_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6425_cast_fp16, var_6664_cast_fp16))[name = string("op_6721_cast_fp16")]; string var_6723_equation_0 = const()[name = string("op_6723_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6723_cast_fp16 = einsum(equation = var_6723_equation_0, values = (var_6425_cast_fp16, var_6665_cast_fp16))[name = string("op_6723_cast_fp16")]; string var_6725_equation_0 = const()[name = string("op_6725_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6425_cast_fp16, var_6666_cast_fp16))[name = string("op_6725_cast_fp16")]; string var_6727_equation_0 = const()[name = string("op_6727_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6727_cast_fp16 = einsum(equation = var_6727_equation_0, values = (var_6425_cast_fp16, var_6667_cast_fp16))[name = string("op_6727_cast_fp16")]; string var_6729_equation_0 = const()[name = string("op_6729_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6429_cast_fp16, var_6668_cast_fp16))[name = string("op_6729_cast_fp16")]; string var_6731_equation_0 = const()[name = string("op_6731_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6731_cast_fp16 = einsum(equation = var_6731_equation_0, values = (var_6429_cast_fp16, var_6669_cast_fp16))[name = string("op_6731_cast_fp16")]; string var_6733_equation_0 = const()[name = string("op_6733_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6733_cast_fp16 = einsum(equation = var_6733_equation_0, values = (var_6429_cast_fp16, var_6670_cast_fp16))[name = string("op_6733_cast_fp16")]; string var_6735_equation_0 = const()[name = string("op_6735_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6735_cast_fp16 = einsum(equation = var_6735_equation_0, values = (var_6429_cast_fp16, var_6671_cast_fp16))[name = string("op_6735_cast_fp16")]; string var_6737_equation_0 = const()[name = string("op_6737_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6737_cast_fp16 = einsum(equation = var_6737_equation_0, values = (var_6433_cast_fp16, var_6672_cast_fp16))[name = string("op_6737_cast_fp16")]; string var_6739_equation_0 = const()[name = string("op_6739_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6739_cast_fp16 = einsum(equation = var_6739_equation_0, values = (var_6433_cast_fp16, var_6673_cast_fp16))[name = string("op_6739_cast_fp16")]; string var_6741_equation_0 = const()[name = string("op_6741_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6741_cast_fp16 = einsum(equation = var_6741_equation_0, values = (var_6433_cast_fp16, var_6674_cast_fp16))[name = string("op_6741_cast_fp16")]; string var_6743_equation_0 = const()[name = string("op_6743_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6743_cast_fp16 = einsum(equation = var_6743_equation_0, values = (var_6433_cast_fp16, var_6675_cast_fp16))[name = string("op_6743_cast_fp16")]; string var_6745_equation_0 = const()[name = string("op_6745_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6745_cast_fp16 = einsum(equation = var_6745_equation_0, values = (var_6437_cast_fp16, var_6676_cast_fp16))[name = string("op_6745_cast_fp16")]; string var_6747_equation_0 = const()[name = string("op_6747_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6747_cast_fp16 = einsum(equation = var_6747_equation_0, values = (var_6437_cast_fp16, var_6677_cast_fp16))[name = string("op_6747_cast_fp16")]; string var_6749_equation_0 = const()[name = string("op_6749_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6749_cast_fp16 = einsum(equation = var_6749_equation_0, values = (var_6437_cast_fp16, var_6678_cast_fp16))[name = string("op_6749_cast_fp16")]; string var_6751_equation_0 = const()[name = string("op_6751_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6751_cast_fp16 = einsum(equation = var_6751_equation_0, values = (var_6437_cast_fp16, var_6679_cast_fp16))[name = string("op_6751_cast_fp16")]; string var_6753_equation_0 = const()[name = string("op_6753_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6441_cast_fp16, var_6680_cast_fp16))[name = string("op_6753_cast_fp16")]; string var_6755_equation_0 = const()[name = string("op_6755_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6441_cast_fp16, var_6681_cast_fp16))[name = string("op_6755_cast_fp16")]; string var_6757_equation_0 = const()[name = string("op_6757_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6441_cast_fp16, var_6682_cast_fp16))[name = string("op_6757_cast_fp16")]; string var_6759_equation_0 = const()[name = string("op_6759_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6441_cast_fp16, var_6683_cast_fp16))[name = string("op_6759_cast_fp16")]; string var_6761_equation_0 = const()[name = string("op_6761_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6445_cast_fp16, var_6684_cast_fp16))[name = string("op_6761_cast_fp16")]; string var_6763_equation_0 = const()[name = string("op_6763_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6445_cast_fp16, var_6685_cast_fp16))[name = string("op_6763_cast_fp16")]; string var_6765_equation_0 = const()[name = string("op_6765_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6445_cast_fp16, var_6686_cast_fp16))[name = string("op_6765_cast_fp16")]; string var_6767_equation_0 = const()[name = string("op_6767_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6445_cast_fp16, var_6687_cast_fp16))[name = string("op_6767_cast_fp16")]; string var_6769_equation_0 = const()[name = string("op_6769_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6449_cast_fp16, var_6688_cast_fp16))[name = string("op_6769_cast_fp16")]; string var_6771_equation_0 = const()[name = string("op_6771_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6449_cast_fp16, var_6689_cast_fp16))[name = string("op_6771_cast_fp16")]; string var_6773_equation_0 = const()[name = string("op_6773_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6449_cast_fp16, var_6690_cast_fp16))[name = string("op_6773_cast_fp16")]; string var_6775_equation_0 = const()[name = string("op_6775_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6449_cast_fp16, var_6691_cast_fp16))[name = string("op_6775_cast_fp16")]; string var_6777_equation_0 = const()[name = string("op_6777_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6453_cast_fp16, var_6692_cast_fp16))[name = string("op_6777_cast_fp16")]; string var_6779_equation_0 = const()[name = string("op_6779_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6779_cast_fp16 = einsum(equation = var_6779_equation_0, values = (var_6453_cast_fp16, var_6693_cast_fp16))[name = string("op_6779_cast_fp16")]; string var_6781_equation_0 = const()[name = string("op_6781_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6453_cast_fp16, var_6694_cast_fp16))[name = string("op_6781_cast_fp16")]; string var_6783_equation_0 = const()[name = string("op_6783_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6783_cast_fp16 = einsum(equation = var_6783_equation_0, values = (var_6453_cast_fp16, var_6695_cast_fp16))[name = string("op_6783_cast_fp16")]; string var_6785_equation_0 = const()[name = string("op_6785_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6785_cast_fp16 = einsum(equation = var_6785_equation_0, values = (var_6457_cast_fp16, var_6696_cast_fp16))[name = string("op_6785_cast_fp16")]; string var_6787_equation_0 = const()[name = string("op_6787_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6787_cast_fp16 = einsum(equation = var_6787_equation_0, values = (var_6457_cast_fp16, var_6697_cast_fp16))[name = string("op_6787_cast_fp16")]; string var_6789_equation_0 = const()[name = string("op_6789_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6789_cast_fp16 = einsum(equation = var_6789_equation_0, values = (var_6457_cast_fp16, var_6698_cast_fp16))[name = string("op_6789_cast_fp16")]; string var_6791_equation_0 = const()[name = string("op_6791_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6791_cast_fp16 = einsum(equation = var_6791_equation_0, values = (var_6457_cast_fp16, var_6699_cast_fp16))[name = string("op_6791_cast_fp16")]; string var_6793_equation_0 = const()[name = string("op_6793_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6793_cast_fp16 = einsum(equation = var_6793_equation_0, values = (var_6461_cast_fp16, var_6700_cast_fp16))[name = string("op_6793_cast_fp16")]; string var_6795_equation_0 = const()[name = string("op_6795_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6795_cast_fp16 = einsum(equation = var_6795_equation_0, values = (var_6461_cast_fp16, var_6701_cast_fp16))[name = string("op_6795_cast_fp16")]; string var_6797_equation_0 = const()[name = string("op_6797_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6797_cast_fp16 = einsum(equation = var_6797_equation_0, values = (var_6461_cast_fp16, var_6702_cast_fp16))[name = string("op_6797_cast_fp16")]; string var_6799_equation_0 = const()[name = string("op_6799_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_6799_cast_fp16 = einsum(equation = var_6799_equation_0, values = (var_6461_cast_fp16, var_6703_cast_fp16))[name = string("op_6799_cast_fp16")]; bool var_6801_interleave_0 = const()[name = string("op_6801_interleave_0"), val = bool(false)]; tensor var_6801_cast_fp16 = concat(axis = var_5912, interleave = var_6801_interleave_0, values = (var_6705_cast_fp16, var_6707_cast_fp16, var_6709_cast_fp16, var_6711_cast_fp16))[name = string("op_6801_cast_fp16")]; bool var_6803_interleave_0 = const()[name = string("op_6803_interleave_0"), val = bool(false)]; tensor var_6803_cast_fp16 = concat(axis = var_5912, interleave = var_6803_interleave_0, values = (var_6713_cast_fp16, var_6715_cast_fp16, var_6717_cast_fp16, var_6719_cast_fp16))[name = string("op_6803_cast_fp16")]; bool var_6805_interleave_0 = const()[name = string("op_6805_interleave_0"), val = bool(false)]; tensor var_6805_cast_fp16 = concat(axis = var_5912, interleave = var_6805_interleave_0, values = (var_6721_cast_fp16, var_6723_cast_fp16, var_6725_cast_fp16, var_6727_cast_fp16))[name = string("op_6805_cast_fp16")]; bool var_6807_interleave_0 = const()[name = string("op_6807_interleave_0"), val = bool(false)]; tensor var_6807_cast_fp16 = concat(axis = var_5912, interleave = var_6807_interleave_0, values = (var_6729_cast_fp16, var_6731_cast_fp16, var_6733_cast_fp16, var_6735_cast_fp16))[name = string("op_6807_cast_fp16")]; bool var_6809_interleave_0 = const()[name = string("op_6809_interleave_0"), val = bool(false)]; tensor var_6809_cast_fp16 = concat(axis = var_5912, interleave = var_6809_interleave_0, values = (var_6737_cast_fp16, var_6739_cast_fp16, var_6741_cast_fp16, var_6743_cast_fp16))[name = string("op_6809_cast_fp16")]; bool var_6811_interleave_0 = const()[name = string("op_6811_interleave_0"), val = bool(false)]; tensor var_6811_cast_fp16 = concat(axis = var_5912, interleave = var_6811_interleave_0, values = (var_6745_cast_fp16, var_6747_cast_fp16, var_6749_cast_fp16, var_6751_cast_fp16))[name = string("op_6811_cast_fp16")]; bool var_6813_interleave_0 = const()[name = string("op_6813_interleave_0"), val = bool(false)]; tensor var_6813_cast_fp16 = concat(axis = var_5912, interleave = var_6813_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16))[name = string("op_6813_cast_fp16")]; bool var_6815_interleave_0 = const()[name = string("op_6815_interleave_0"), val = bool(false)]; tensor var_6815_cast_fp16 = concat(axis = var_5912, interleave = var_6815_interleave_0, values = (var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16))[name = string("op_6815_cast_fp16")]; bool var_6817_interleave_0 = const()[name = string("op_6817_interleave_0"), val = bool(false)]; tensor var_6817_cast_fp16 = concat(axis = var_5912, interleave = var_6817_interleave_0, values = (var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16))[name = string("op_6817_cast_fp16")]; bool var_6819_interleave_0 = const()[name = string("op_6819_interleave_0"), val = bool(false)]; tensor var_6819_cast_fp16 = concat(axis = var_5912, interleave = var_6819_interleave_0, values = (var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16))[name = string("op_6819_cast_fp16")]; bool var_6821_interleave_0 = const()[name = string("op_6821_interleave_0"), val = bool(false)]; tensor var_6821_cast_fp16 = concat(axis = var_5912, interleave = var_6821_interleave_0, values = (var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16))[name = string("op_6821_cast_fp16")]; bool var_6823_interleave_0 = const()[name = string("op_6823_interleave_0"), val = bool(false)]; tensor var_6823_cast_fp16 = concat(axis = var_5912, interleave = var_6823_interleave_0, values = (var_6793_cast_fp16, var_6795_cast_fp16, var_6797_cast_fp16, var_6799_cast_fp16))[name = string("op_6823_cast_fp16")]; bool input_49_interleave_0 = const()[name = string("input_49_interleave_0"), val = bool(false)]; tensor input_49_cast_fp16 = concat(axis = var_5929, interleave = input_49_interleave_0, values = (var_6801_cast_fp16, var_6803_cast_fp16, var_6805_cast_fp16, var_6807_cast_fp16, var_6809_cast_fp16, var_6811_cast_fp16, var_6813_cast_fp16, var_6815_cast_fp16, var_6817_cast_fp16, var_6819_cast_fp16, var_6821_cast_fp16, var_6823_cast_fp16))[name = string("input_49_cast_fp16")]; string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94814784)))]; tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95994496)))]; tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; fp16 var_6842_to_fp16 = const()[name = string("op_6842_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_6842_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95996096)))]; tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95997696)))]; fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95999296)))]; tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100717952)))]; tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100724160)))]; tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105442816)))]; tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; int32 var_6871 = const()[name = string("op_6871"), val = int32(3)]; int32 var_6888 = const()[name = string("op_6888"), val = int32(1)]; tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; fp16 var_6905_to_fp16 = const()[name = string("op_6905_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_6905_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105444416)))]; tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105446016)))]; fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105447616)))]; tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106627328)))]; tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106628928)))]; tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107808640)))]; tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108988352)))]; tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; tensor var_6943_begin_0 = const()[name = string("op_6943_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6943_end_0 = const()[name = string("op_6943_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6943_end_mask_0 = const()[name = string("op_6943_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = query_15_cast_fp16)[name = string("op_6943_cast_fp16")]; tensor var_6947_begin_0 = const()[name = string("op_6947_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6947_end_0 = const()[name = string("op_6947_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_6947_end_mask_0 = const()[name = string("op_6947_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = query_15_cast_fp16)[name = string("op_6947_cast_fp16")]; tensor var_6951_begin_0 = const()[name = string("op_6951_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6951_end_0 = const()[name = string("op_6951_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_6951_end_mask_0 = const()[name = string("op_6951_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = query_15_cast_fp16)[name = string("op_6951_cast_fp16")]; tensor var_6955_begin_0 = const()[name = string("op_6955_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6955_end_0 = const()[name = string("op_6955_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_6955_end_mask_0 = const()[name = string("op_6955_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = query_15_cast_fp16)[name = string("op_6955_cast_fp16")]; tensor var_6959_begin_0 = const()[name = string("op_6959_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6959_end_0 = const()[name = string("op_6959_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_6959_end_mask_0 = const()[name = string("op_6959_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = query_15_cast_fp16)[name = string("op_6959_cast_fp16")]; tensor var_6963_begin_0 = const()[name = string("op_6963_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6963_end_0 = const()[name = string("op_6963_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_6963_end_mask_0 = const()[name = string("op_6963_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = query_15_cast_fp16)[name = string("op_6963_cast_fp16")]; tensor var_6967_begin_0 = const()[name = string("op_6967_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6967_end_0 = const()[name = string("op_6967_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_6967_end_mask_0 = const()[name = string("op_6967_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6967_cast_fp16 = slice_by_index(begin = var_6967_begin_0, end = var_6967_end_0, end_mask = var_6967_end_mask_0, x = query_15_cast_fp16)[name = string("op_6967_cast_fp16")]; tensor var_6971_begin_0 = const()[name = string("op_6971_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6971_end_0 = const()[name = string("op_6971_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_6971_end_mask_0 = const()[name = string("op_6971_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6971_cast_fp16 = slice_by_index(begin = var_6971_begin_0, end = var_6971_end_0, end_mask = var_6971_end_mask_0, x = query_15_cast_fp16)[name = string("op_6971_cast_fp16")]; tensor var_6975_begin_0 = const()[name = string("op_6975_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6975_end_0 = const()[name = string("op_6975_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_6975_end_mask_0 = const()[name = string("op_6975_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = query_15_cast_fp16)[name = string("op_6975_cast_fp16")]; tensor var_6979_begin_0 = const()[name = string("op_6979_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6979_end_0 = const()[name = string("op_6979_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_6979_end_mask_0 = const()[name = string("op_6979_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = query_15_cast_fp16)[name = string("op_6979_cast_fp16")]; tensor var_6983_begin_0 = const()[name = string("op_6983_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6983_end_0 = const()[name = string("op_6983_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6983_end_mask_0 = const()[name = string("op_6983_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = query_15_cast_fp16)[name = string("op_6983_cast_fp16")]; tensor var_6987_begin_0 = const()[name = string("op_6987_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6987_end_0 = const()[name = string("op_6987_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6987_end_mask_0 = const()[name = string("op_6987_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = query_15_cast_fp16)[name = string("op_6987_cast_fp16")]; tensor var_6996_begin_0 = const()[name = string("op_6996_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6996_end_0 = const()[name = string("op_6996_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6996_end_mask_0 = const()[name = string("op_6996_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6996_cast_fp16 = slice_by_index(begin = var_6996_begin_0, end = var_6996_end_0, end_mask = var_6996_end_mask_0, x = var_6943_cast_fp16)[name = string("op_6996_cast_fp16")]; tensor var_7003_begin_0 = const()[name = string("op_7003_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7003_end_0 = const()[name = string("op_7003_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7003_end_mask_0 = const()[name = string("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7003_cast_fp16")]; tensor var_7010_begin_0 = const()[name = string("op_7010_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7010_end_0 = const()[name = string("op_7010_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7010_end_mask_0 = const()[name = string("op_7010_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7010_cast_fp16 = slice_by_index(begin = var_7010_begin_0, end = var_7010_end_0, end_mask = var_7010_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7010_cast_fp16")]; tensor var_7017_begin_0 = const()[name = string("op_7017_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7017_end_0 = const()[name = string("op_7017_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7017_end_mask_0 = const()[name = string("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7017_cast_fp16")]; tensor var_7024_begin_0 = const()[name = string("op_7024_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7024_end_0 = const()[name = string("op_7024_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7024_end_mask_0 = const()[name = string("op_7024_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7024_cast_fp16 = slice_by_index(begin = var_7024_begin_0, end = var_7024_end_0, end_mask = var_7024_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7024_cast_fp16")]; tensor var_7031_begin_0 = const()[name = string("op_7031_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7031_end_0 = const()[name = string("op_7031_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7031_end_mask_0 = const()[name = string("op_7031_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7031_cast_fp16 = slice_by_index(begin = var_7031_begin_0, end = var_7031_end_0, end_mask = var_7031_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7031_cast_fp16")]; tensor var_7038_begin_0 = const()[name = string("op_7038_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7038_end_0 = const()[name = string("op_7038_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7038_end_mask_0 = const()[name = string("op_7038_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7038_cast_fp16 = slice_by_index(begin = var_7038_begin_0, end = var_7038_end_0, end_mask = var_7038_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7038_cast_fp16")]; tensor var_7045_begin_0 = const()[name = string("op_7045_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7045_end_0 = const()[name = string("op_7045_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7045_end_mask_0 = const()[name = string("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7045_cast_fp16")]; tensor var_7052_begin_0 = const()[name = string("op_7052_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7052_end_0 = const()[name = string("op_7052_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7052_end_mask_0 = const()[name = string("op_7052_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7052_cast_fp16 = slice_by_index(begin = var_7052_begin_0, end = var_7052_end_0, end_mask = var_7052_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7052_cast_fp16")]; tensor var_7059_begin_0 = const()[name = string("op_7059_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7059_end_0 = const()[name = string("op_7059_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7059_end_mask_0 = const()[name = string("op_7059_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7059_cast_fp16 = slice_by_index(begin = var_7059_begin_0, end = var_7059_end_0, end_mask = var_7059_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7059_cast_fp16")]; tensor var_7066_begin_0 = const()[name = string("op_7066_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7066_end_0 = const()[name = string("op_7066_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7066_end_mask_0 = const()[name = string("op_7066_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7066_cast_fp16 = slice_by_index(begin = var_7066_begin_0, end = var_7066_end_0, end_mask = var_7066_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7066_cast_fp16")]; tensor var_7073_begin_0 = const()[name = string("op_7073_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7073_end_0 = const()[name = string("op_7073_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7073_end_mask_0 = const()[name = string("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7073_cast_fp16")]; tensor var_7080_begin_0 = const()[name = string("op_7080_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7080_end_0 = const()[name = string("op_7080_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7080_end_mask_0 = const()[name = string("op_7080_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7080_cast_fp16 = slice_by_index(begin = var_7080_begin_0, end = var_7080_end_0, end_mask = var_7080_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7080_cast_fp16")]; tensor var_7087_begin_0 = const()[name = string("op_7087_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7087_end_0 = const()[name = string("op_7087_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7087_end_mask_0 = const()[name = string("op_7087_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7087_cast_fp16 = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7087_cast_fp16")]; tensor var_7094_begin_0 = const()[name = string("op_7094_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7094_end_0 = const()[name = string("op_7094_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7094_end_mask_0 = const()[name = string("op_7094_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7094_cast_fp16 = slice_by_index(begin = var_7094_begin_0, end = var_7094_end_0, end_mask = var_7094_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7094_cast_fp16")]; tensor var_7101_begin_0 = const()[name = string("op_7101_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7101_end_0 = const()[name = string("op_7101_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7101_end_mask_0 = const()[name = string("op_7101_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7101_cast_fp16 = slice_by_index(begin = var_7101_begin_0, end = var_7101_end_0, end_mask = var_7101_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7101_cast_fp16")]; tensor var_7108_begin_0 = const()[name = string("op_7108_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7108_end_0 = const()[name = string("op_7108_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7108_end_mask_0 = const()[name = string("op_7108_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7108_cast_fp16 = slice_by_index(begin = var_7108_begin_0, end = var_7108_end_0, end_mask = var_7108_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7108_cast_fp16")]; tensor var_7115_begin_0 = const()[name = string("op_7115_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7115_end_0 = const()[name = string("op_7115_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7115_end_mask_0 = const()[name = string("op_7115_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7115_cast_fp16")]; tensor var_7122_begin_0 = const()[name = string("op_7122_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7122_end_0 = const()[name = string("op_7122_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7122_end_mask_0 = const()[name = string("op_7122_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7122_cast_fp16 = slice_by_index(begin = var_7122_begin_0, end = var_7122_end_0, end_mask = var_7122_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7122_cast_fp16")]; tensor var_7129_begin_0 = const()[name = string("op_7129_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7129_end_0 = const()[name = string("op_7129_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7129_end_mask_0 = const()[name = string("op_7129_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7129_cast_fp16 = slice_by_index(begin = var_7129_begin_0, end = var_7129_end_0, end_mask = var_7129_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7129_cast_fp16")]; tensor var_7136_begin_0 = const()[name = string("op_7136_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7136_end_0 = const()[name = string("op_7136_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7136_end_mask_0 = const()[name = string("op_7136_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7136_cast_fp16 = slice_by_index(begin = var_7136_begin_0, end = var_7136_end_0, end_mask = var_7136_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7136_cast_fp16")]; tensor var_7143_begin_0 = const()[name = string("op_7143_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7143_end_0 = const()[name = string("op_7143_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7143_end_mask_0 = const()[name = string("op_7143_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7143_cast_fp16")]; tensor var_7150_begin_0 = const()[name = string("op_7150_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7150_end_0 = const()[name = string("op_7150_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7150_end_mask_0 = const()[name = string("op_7150_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7150_cast_fp16 = slice_by_index(begin = var_7150_begin_0, end = var_7150_end_0, end_mask = var_7150_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7150_cast_fp16")]; tensor var_7157_begin_0 = const()[name = string("op_7157_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7157_end_0 = const()[name = string("op_7157_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7157_end_mask_0 = const()[name = string("op_7157_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7157_cast_fp16 = slice_by_index(begin = var_7157_begin_0, end = var_7157_end_0, end_mask = var_7157_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7157_cast_fp16")]; tensor var_7164_begin_0 = const()[name = string("op_7164_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7164_end_0 = const()[name = string("op_7164_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7164_end_mask_0 = const()[name = string("op_7164_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7164_cast_fp16 = slice_by_index(begin = var_7164_begin_0, end = var_7164_end_0, end_mask = var_7164_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7164_cast_fp16")]; tensor var_7171_begin_0 = const()[name = string("op_7171_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7171_end_0 = const()[name = string("op_7171_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7171_end_mask_0 = const()[name = string("op_7171_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7171_cast_fp16 = slice_by_index(begin = var_7171_begin_0, end = var_7171_end_0, end_mask = var_7171_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7171_cast_fp16")]; tensor var_7178_begin_0 = const()[name = string("op_7178_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7178_end_0 = const()[name = string("op_7178_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7178_end_mask_0 = const()[name = string("op_7178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7178_cast_fp16")]; tensor var_7185_begin_0 = const()[name = string("op_7185_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7185_end_0 = const()[name = string("op_7185_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7185_end_mask_0 = const()[name = string("op_7185_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7185_cast_fp16 = slice_by_index(begin = var_7185_begin_0, end = var_7185_end_0, end_mask = var_7185_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7185_cast_fp16")]; tensor var_7192_begin_0 = const()[name = string("op_7192_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7192_end_0 = const()[name = string("op_7192_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7192_end_mask_0 = const()[name = string("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7192_cast_fp16")]; tensor var_7199_begin_0 = const()[name = string("op_7199_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7199_end_0 = const()[name = string("op_7199_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7199_end_mask_0 = const()[name = string("op_7199_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7199_cast_fp16 = slice_by_index(begin = var_7199_begin_0, end = var_7199_end_0, end_mask = var_7199_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7199_cast_fp16")]; tensor var_7206_begin_0 = const()[name = string("op_7206_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7206_end_0 = const()[name = string("op_7206_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7206_end_mask_0 = const()[name = string("op_7206_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7206_cast_fp16")]; tensor var_7213_begin_0 = const()[name = string("op_7213_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7213_end_0 = const()[name = string("op_7213_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7213_end_mask_0 = const()[name = string("op_7213_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7213_cast_fp16")]; tensor var_7220_begin_0 = const()[name = string("op_7220_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7220_end_0 = const()[name = string("op_7220_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7220_end_mask_0 = const()[name = string("op_7220_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7220_cast_fp16 = slice_by_index(begin = var_7220_begin_0, end = var_7220_end_0, end_mask = var_7220_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7220_cast_fp16")]; tensor var_7227_begin_0 = const()[name = string("op_7227_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7227_end_0 = const()[name = string("op_7227_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7227_end_mask_0 = const()[name = string("op_7227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7227_cast_fp16 = slice_by_index(begin = var_7227_begin_0, end = var_7227_end_0, end_mask = var_7227_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7227_cast_fp16")]; tensor var_7234_begin_0 = const()[name = string("op_7234_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7234_end_0 = const()[name = string("op_7234_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7234_end_mask_0 = const()[name = string("op_7234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7234_cast_fp16")]; tensor var_7241_begin_0 = const()[name = string("op_7241_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7241_end_0 = const()[name = string("op_7241_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7241_end_mask_0 = const()[name = string("op_7241_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7241_cast_fp16 = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7241_cast_fp16")]; tensor var_7248_begin_0 = const()[name = string("op_7248_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7248_end_0 = const()[name = string("op_7248_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7248_end_mask_0 = const()[name = string("op_7248_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7248_cast_fp16 = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7248_cast_fp16")]; tensor var_7255_begin_0 = const()[name = string("op_7255_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7255_end_0 = const()[name = string("op_7255_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7255_end_mask_0 = const()[name = string("op_7255_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7255_cast_fp16 = slice_by_index(begin = var_7255_begin_0, end = var_7255_end_0, end_mask = var_7255_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7255_cast_fp16")]; tensor var_7262_begin_0 = const()[name = string("op_7262_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7262_end_0 = const()[name = string("op_7262_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7262_end_mask_0 = const()[name = string("op_7262_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7262_cast_fp16 = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7262_cast_fp16")]; tensor var_7269_begin_0 = const()[name = string("op_7269_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7269_end_0 = const()[name = string("op_7269_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7269_end_mask_0 = const()[name = string("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7269_cast_fp16")]; tensor var_7276_begin_0 = const()[name = string("op_7276_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7276_end_0 = const()[name = string("op_7276_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7276_end_mask_0 = const()[name = string("op_7276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7276_cast_fp16 = slice_by_index(begin = var_7276_begin_0, end = var_7276_end_0, end_mask = var_7276_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7276_cast_fp16")]; tensor var_7283_begin_0 = const()[name = string("op_7283_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7283_end_0 = const()[name = string("op_7283_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7283_end_mask_0 = const()[name = string("op_7283_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7283_cast_fp16 = slice_by_index(begin = var_7283_begin_0, end = var_7283_end_0, end_mask = var_7283_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7283_cast_fp16")]; tensor var_7290_begin_0 = const()[name = string("op_7290_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7290_end_0 = const()[name = string("op_7290_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7290_end_mask_0 = const()[name = string("op_7290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7290_cast_fp16 = slice_by_index(begin = var_7290_begin_0, end = var_7290_end_0, end_mask = var_7290_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7290_cast_fp16")]; tensor var_7297_begin_0 = const()[name = string("op_7297_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7297_end_0 = const()[name = string("op_7297_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7297_end_mask_0 = const()[name = string("op_7297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7297_cast_fp16 = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7297_cast_fp16")]; tensor var_7304_begin_0 = const()[name = string("op_7304_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7304_end_0 = const()[name = string("op_7304_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7304_end_mask_0 = const()[name = string("op_7304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7304_cast_fp16 = slice_by_index(begin = var_7304_begin_0, end = var_7304_end_0, end_mask = var_7304_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7304_cast_fp16")]; tensor var_7311_begin_0 = const()[name = string("op_7311_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7311_end_0 = const()[name = string("op_7311_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7311_end_mask_0 = const()[name = string("op_7311_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7311_cast_fp16 = slice_by_index(begin = var_7311_begin_0, end = var_7311_end_0, end_mask = var_7311_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7311_cast_fp16")]; tensor var_7318_begin_0 = const()[name = string("op_7318_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7318_end_0 = const()[name = string("op_7318_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7318_end_mask_0 = const()[name = string("op_7318_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7318_cast_fp16 = slice_by_index(begin = var_7318_begin_0, end = var_7318_end_0, end_mask = var_7318_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7318_cast_fp16")]; tensor var_7325_begin_0 = const()[name = string("op_7325_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7325_end_0 = const()[name = string("op_7325_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7325_end_mask_0 = const()[name = string("op_7325_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7325_cast_fp16 = slice_by_index(begin = var_7325_begin_0, end = var_7325_end_0, end_mask = var_7325_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7325_cast_fp16")]; tensor k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7330_begin_0 = const()[name = string("op_7330_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7330_end_0 = const()[name = string("op_7330_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_7330_end_mask_0 = const()[name = string("op_7330_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = string("transpose_4")]; tensor var_7330_cast_fp16 = slice_by_index(begin = var_7330_begin_0, end = var_7330_end_0, end_mask = var_7330_end_mask_0, x = k_15_cast_fp16)[name = string("op_7330_cast_fp16")]; tensor var_7334_begin_0 = const()[name = string("op_7334_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_7334_end_0 = const()[name = string("op_7334_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_7334_end_mask_0 = const()[name = string("op_7334_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7334_cast_fp16 = slice_by_index(begin = var_7334_begin_0, end = var_7334_end_0, end_mask = var_7334_end_mask_0, x = k_15_cast_fp16)[name = string("op_7334_cast_fp16")]; tensor var_7338_begin_0 = const()[name = string("op_7338_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_7338_end_0 = const()[name = string("op_7338_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_7338_end_mask_0 = const()[name = string("op_7338_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7338_cast_fp16 = slice_by_index(begin = var_7338_begin_0, end = var_7338_end_0, end_mask = var_7338_end_mask_0, x = k_15_cast_fp16)[name = string("op_7338_cast_fp16")]; tensor var_7342_begin_0 = const()[name = string("op_7342_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_7342_end_0 = const()[name = string("op_7342_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_7342_end_mask_0 = const()[name = string("op_7342_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7342_cast_fp16 = slice_by_index(begin = var_7342_begin_0, end = var_7342_end_0, end_mask = var_7342_end_mask_0, x = k_15_cast_fp16)[name = string("op_7342_cast_fp16")]; tensor var_7346_begin_0 = const()[name = string("op_7346_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_7346_end_0 = const()[name = string("op_7346_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_7346_end_mask_0 = const()[name = string("op_7346_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7346_cast_fp16 = slice_by_index(begin = var_7346_begin_0, end = var_7346_end_0, end_mask = var_7346_end_mask_0, x = k_15_cast_fp16)[name = string("op_7346_cast_fp16")]; tensor var_7350_begin_0 = const()[name = string("op_7350_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7350_end_0 = const()[name = string("op_7350_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_7350_end_mask_0 = const()[name = string("op_7350_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7350_cast_fp16 = slice_by_index(begin = var_7350_begin_0, end = var_7350_end_0, end_mask = var_7350_end_mask_0, x = k_15_cast_fp16)[name = string("op_7350_cast_fp16")]; tensor var_7354_begin_0 = const()[name = string("op_7354_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_7354_end_0 = const()[name = string("op_7354_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_7354_end_mask_0 = const()[name = string("op_7354_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7354_cast_fp16 = slice_by_index(begin = var_7354_begin_0, end = var_7354_end_0, end_mask = var_7354_end_mask_0, x = k_15_cast_fp16)[name = string("op_7354_cast_fp16")]; tensor var_7358_begin_0 = const()[name = string("op_7358_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_7358_end_0 = const()[name = string("op_7358_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_7358_end_mask_0 = const()[name = string("op_7358_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7358_cast_fp16 = slice_by_index(begin = var_7358_begin_0, end = var_7358_end_0, end_mask = var_7358_end_mask_0, x = k_15_cast_fp16)[name = string("op_7358_cast_fp16")]; tensor var_7362_begin_0 = const()[name = string("op_7362_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7362_end_0 = const()[name = string("op_7362_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_7362_end_mask_0 = const()[name = string("op_7362_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7362_cast_fp16 = slice_by_index(begin = var_7362_begin_0, end = var_7362_end_0, end_mask = var_7362_end_mask_0, x = k_15_cast_fp16)[name = string("op_7362_cast_fp16")]; tensor var_7366_begin_0 = const()[name = string("op_7366_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_7366_end_0 = const()[name = string("op_7366_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_7366_end_mask_0 = const()[name = string("op_7366_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7366_cast_fp16 = slice_by_index(begin = var_7366_begin_0, end = var_7366_end_0, end_mask = var_7366_end_mask_0, x = k_15_cast_fp16)[name = string("op_7366_cast_fp16")]; tensor var_7370_begin_0 = const()[name = string("op_7370_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7370_end_0 = const()[name = string("op_7370_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_7370_end_mask_0 = const()[name = string("op_7370_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7370_cast_fp16 = slice_by_index(begin = var_7370_begin_0, end = var_7370_end_0, end_mask = var_7370_end_mask_0, x = k_15_cast_fp16)[name = string("op_7370_cast_fp16")]; tensor var_7374_begin_0 = const()[name = string("op_7374_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_7374_end_0 = const()[name = string("op_7374_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_7374_end_mask_0 = const()[name = string("op_7374_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7374_cast_fp16 = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = k_15_cast_fp16)[name = string("op_7374_cast_fp16")]; tensor var_7376_begin_0 = const()[name = string("op_7376_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7376_end_0 = const()[name = string("op_7376_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7376_end_mask_0 = const()[name = string("op_7376_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7376_cast_fp16 = slice_by_index(begin = var_7376_begin_0, end = var_7376_end_0, end_mask = var_7376_end_mask_0, x = value_15_cast_fp16)[name = string("op_7376_cast_fp16")]; tensor var_7380_begin_0 = const()[name = string("op_7380_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7380_end_0 = const()[name = string("op_7380_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_7380_end_mask_0 = const()[name = string("op_7380_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7380_cast_fp16 = slice_by_index(begin = var_7380_begin_0, end = var_7380_end_0, end_mask = var_7380_end_mask_0, x = value_15_cast_fp16)[name = string("op_7380_cast_fp16")]; tensor var_7384_begin_0 = const()[name = string("op_7384_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7384_end_0 = const()[name = string("op_7384_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_7384_end_mask_0 = const()[name = string("op_7384_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7384_cast_fp16 = slice_by_index(begin = var_7384_begin_0, end = var_7384_end_0, end_mask = var_7384_end_mask_0, x = value_15_cast_fp16)[name = string("op_7384_cast_fp16")]; tensor var_7388_begin_0 = const()[name = string("op_7388_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7388_end_0 = const()[name = string("op_7388_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_7388_end_mask_0 = const()[name = string("op_7388_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7388_cast_fp16 = slice_by_index(begin = var_7388_begin_0, end = var_7388_end_0, end_mask = var_7388_end_mask_0, x = value_15_cast_fp16)[name = string("op_7388_cast_fp16")]; tensor var_7392_begin_0 = const()[name = string("op_7392_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7392_end_0 = const()[name = string("op_7392_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_7392_end_mask_0 = const()[name = string("op_7392_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7392_cast_fp16 = slice_by_index(begin = var_7392_begin_0, end = var_7392_end_0, end_mask = var_7392_end_mask_0, x = value_15_cast_fp16)[name = string("op_7392_cast_fp16")]; tensor var_7396_begin_0 = const()[name = string("op_7396_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7396_end_0 = const()[name = string("op_7396_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_7396_end_mask_0 = const()[name = string("op_7396_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7396_cast_fp16 = slice_by_index(begin = var_7396_begin_0, end = var_7396_end_0, end_mask = var_7396_end_mask_0, x = value_15_cast_fp16)[name = string("op_7396_cast_fp16")]; tensor var_7400_begin_0 = const()[name = string("op_7400_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7400_end_0 = const()[name = string("op_7400_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_7400_end_mask_0 = const()[name = string("op_7400_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7400_cast_fp16 = slice_by_index(begin = var_7400_begin_0, end = var_7400_end_0, end_mask = var_7400_end_mask_0, x = value_15_cast_fp16)[name = string("op_7400_cast_fp16")]; tensor var_7404_begin_0 = const()[name = string("op_7404_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7404_end_0 = const()[name = string("op_7404_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_7404_end_mask_0 = const()[name = string("op_7404_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7404_cast_fp16 = slice_by_index(begin = var_7404_begin_0, end = var_7404_end_0, end_mask = var_7404_end_mask_0, x = value_15_cast_fp16)[name = string("op_7404_cast_fp16")]; tensor var_7408_begin_0 = const()[name = string("op_7408_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7408_end_0 = const()[name = string("op_7408_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_7408_end_mask_0 = const()[name = string("op_7408_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7408_cast_fp16 = slice_by_index(begin = var_7408_begin_0, end = var_7408_end_0, end_mask = var_7408_end_mask_0, x = value_15_cast_fp16)[name = string("op_7408_cast_fp16")]; tensor var_7412_begin_0 = const()[name = string("op_7412_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7412_end_0 = const()[name = string("op_7412_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_7412_end_mask_0 = const()[name = string("op_7412_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7412_cast_fp16 = slice_by_index(begin = var_7412_begin_0, end = var_7412_end_0, end_mask = var_7412_end_mask_0, x = value_15_cast_fp16)[name = string("op_7412_cast_fp16")]; tensor var_7416_begin_0 = const()[name = string("op_7416_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7416_end_0 = const()[name = string("op_7416_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_7416_end_mask_0 = const()[name = string("op_7416_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7416_cast_fp16 = slice_by_index(begin = var_7416_begin_0, end = var_7416_end_0, end_mask = var_7416_end_mask_0, x = value_15_cast_fp16)[name = string("op_7416_cast_fp16")]; tensor var_7420_begin_0 = const()[name = string("op_7420_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7420_end_0 = const()[name = string("op_7420_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_7420_end_mask_0 = const()[name = string("op_7420_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7420_cast_fp16 = slice_by_index(begin = var_7420_begin_0, end = var_7420_end_0, end_mask = var_7420_end_mask_0, x = value_15_cast_fp16)[name = string("op_7420_cast_fp16")]; string _SplitHeadsQ__mh_w_673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_673_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_673_equation_0, values = (var_7330_cast_fp16, var_6996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_673_cast_fp16")]; string _SplitHeadsQ__mh_w_675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_675_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_675_equation_0, values = (var_7330_cast_fp16, var_7003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_675_cast_fp16")]; string _SplitHeadsQ__mh_w_677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_677_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_677_equation_0, values = (var_7330_cast_fp16, var_7010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_677_cast_fp16")]; string _SplitHeadsQ__mh_w_679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_679_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_679_equation_0, values = (var_7330_cast_fp16, var_7017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_679_cast_fp16")]; string _SplitHeadsQ__mh_w_681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_681_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_681_equation_0, values = (var_7334_cast_fp16, var_7024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_681_cast_fp16")]; string _SplitHeadsQ__mh_w_683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_683_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_683_equation_0, values = (var_7334_cast_fp16, var_7031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_683_cast_fp16")]; string _SplitHeadsQ__mh_w_685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_685_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_685_equation_0, values = (var_7334_cast_fp16, var_7038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_685_cast_fp16")]; string _SplitHeadsQ__mh_w_687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_687_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_687_equation_0, values = (var_7334_cast_fp16, var_7045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_687_cast_fp16")]; string _SplitHeadsQ__mh_w_689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_689_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_689_equation_0, values = (var_7338_cast_fp16, var_7052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_689_cast_fp16")]; string _SplitHeadsQ__mh_w_691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_691_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_691_equation_0, values = (var_7338_cast_fp16, var_7059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_691_cast_fp16")]; string _SplitHeadsQ__mh_w_693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_693_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_693_equation_0, values = (var_7338_cast_fp16, var_7066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_693_cast_fp16")]; string _SplitHeadsQ__mh_w_695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_695_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_695_equation_0, values = (var_7338_cast_fp16, var_7073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_695_cast_fp16")]; string _SplitHeadsQ__mh_w_697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_697_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_697_equation_0, values = (var_7342_cast_fp16, var_7080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_697_cast_fp16")]; string _SplitHeadsQ__mh_w_699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_699_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_699_equation_0, values = (var_7342_cast_fp16, var_7087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_699_cast_fp16")]; string _SplitHeadsQ__mh_w_701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_701_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_701_equation_0, values = (var_7342_cast_fp16, var_7094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_701_cast_fp16")]; string _SplitHeadsQ__mh_w_703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_703_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_703_equation_0, values = (var_7342_cast_fp16, var_7101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_703_cast_fp16")]; string _SplitHeadsQ__mh_w_705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_705_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_705_equation_0, values = (var_7346_cast_fp16, var_7108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_705_cast_fp16")]; string _SplitHeadsQ__mh_w_707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_707_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_707_equation_0, values = (var_7346_cast_fp16, var_7115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_707_cast_fp16")]; string _SplitHeadsQ__mh_w_709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_709_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_709_equation_0, values = (var_7346_cast_fp16, var_7122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_709_cast_fp16")]; string _SplitHeadsQ__mh_w_711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_711_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_711_equation_0, values = (var_7346_cast_fp16, var_7129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_711_cast_fp16")]; string _SplitHeadsQ__mh_w_713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_713_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_713_equation_0, values = (var_7350_cast_fp16, var_7136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_713_cast_fp16")]; string _SplitHeadsQ__mh_w_715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_715_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_715_equation_0, values = (var_7350_cast_fp16, var_7143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_715_cast_fp16")]; string _SplitHeadsQ__mh_w_717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_717_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_717_equation_0, values = (var_7350_cast_fp16, var_7150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_717_cast_fp16")]; string _SplitHeadsQ__mh_w_719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_719_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_719_equation_0, values = (var_7350_cast_fp16, var_7157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_719_cast_fp16")]; string _SplitHeadsQ__mh_w_721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_721_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_721_equation_0, values = (var_7354_cast_fp16, var_7164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_721_cast_fp16")]; string _SplitHeadsQ__mh_w_723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_723_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_723_equation_0, values = (var_7354_cast_fp16, var_7171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_723_cast_fp16")]; string _SplitHeadsQ__mh_w_725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_725_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_725_equation_0, values = (var_7354_cast_fp16, var_7178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_725_cast_fp16")]; string _SplitHeadsQ__mh_w_727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_727_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_727_equation_0, values = (var_7354_cast_fp16, var_7185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_727_cast_fp16")]; string _SplitHeadsQ__mh_w_729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_729_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_729_equation_0, values = (var_7358_cast_fp16, var_7192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_729_cast_fp16")]; string _SplitHeadsQ__mh_w_731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_731_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_731_equation_0, values = (var_7358_cast_fp16, var_7199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_731_cast_fp16")]; string _SplitHeadsQ__mh_w_733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_733_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_733_equation_0, values = (var_7358_cast_fp16, var_7206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_733_cast_fp16")]; string _SplitHeadsQ__mh_w_735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_735_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_735_equation_0, values = (var_7358_cast_fp16, var_7213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_735_cast_fp16")]; string _SplitHeadsQ__mh_w_737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_737_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_737_equation_0, values = (var_7362_cast_fp16, var_7220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_737_cast_fp16")]; string _SplitHeadsQ__mh_w_739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_739_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_739_equation_0, values = (var_7362_cast_fp16, var_7227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_739_cast_fp16")]; string _SplitHeadsQ__mh_w_741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_741_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_741_equation_0, values = (var_7362_cast_fp16, var_7234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_741_cast_fp16")]; string _SplitHeadsQ__mh_w_743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_743_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_743_equation_0, values = (var_7362_cast_fp16, var_7241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_743_cast_fp16")]; string _SplitHeadsQ__mh_w_745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_745_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_745_equation_0, values = (var_7366_cast_fp16, var_7248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_745_cast_fp16")]; string _SplitHeadsQ__mh_w_747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_747_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_747_equation_0, values = (var_7366_cast_fp16, var_7255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_747_cast_fp16")]; string _SplitHeadsQ__mh_w_749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_749_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_749_equation_0, values = (var_7366_cast_fp16, var_7262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_749_cast_fp16")]; string _SplitHeadsQ__mh_w_751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_751_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_751_equation_0, values = (var_7366_cast_fp16, var_7269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_751_cast_fp16")]; string _SplitHeadsQ__mh_w_753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_753_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_753_equation_0, values = (var_7370_cast_fp16, var_7276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_753_cast_fp16")]; string _SplitHeadsQ__mh_w_755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_755_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_755_equation_0, values = (var_7370_cast_fp16, var_7283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_755_cast_fp16")]; string _SplitHeadsQ__mh_w_757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_757_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_757_equation_0, values = (var_7370_cast_fp16, var_7290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_757_cast_fp16")]; string _SplitHeadsQ__mh_w_759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_759_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_759_equation_0, values = (var_7370_cast_fp16, var_7297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_759_cast_fp16")]; string _SplitHeadsQ__mh_w_761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_761_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_761_equation_0, values = (var_7374_cast_fp16, var_7304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_761_cast_fp16")]; string _SplitHeadsQ__mh_w_763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_763_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_763_equation_0, values = (var_7374_cast_fp16, var_7311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_763_cast_fp16")]; string _SplitHeadsQ__mh_w_765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_765_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_765_equation_0, values = (var_7374_cast_fp16, var_7318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_765_cast_fp16")]; string _SplitHeadsQ__mh_w_767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_767_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_767_equation_0, values = (var_7374_cast_fp16, var_7325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_767_cast_fp16")]; fp16 var_7519_to_fp16 = const()[name = string("op_7519_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_673_cast_fp16, y = var_7519_to_fp16)[name = string("aw_chunk_673_cast_fp16")]; fp16 var_7521_to_fp16 = const()[name = string("op_7521_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_675_cast_fp16, y = var_7521_to_fp16)[name = string("aw_chunk_675_cast_fp16")]; fp16 var_7523_to_fp16 = const()[name = string("op_7523_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_677_cast_fp16, y = var_7523_to_fp16)[name = string("aw_chunk_677_cast_fp16")]; fp16 var_7525_to_fp16 = const()[name = string("op_7525_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_679_cast_fp16, y = var_7525_to_fp16)[name = string("aw_chunk_679_cast_fp16")]; fp16 var_7527_to_fp16 = const()[name = string("op_7527_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_681_cast_fp16, y = var_7527_to_fp16)[name = string("aw_chunk_681_cast_fp16")]; fp16 var_7529_to_fp16 = const()[name = string("op_7529_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_683_cast_fp16, y = var_7529_to_fp16)[name = string("aw_chunk_683_cast_fp16")]; fp16 var_7531_to_fp16 = const()[name = string("op_7531_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_685_cast_fp16, y = var_7531_to_fp16)[name = string("aw_chunk_685_cast_fp16")]; fp16 var_7533_to_fp16 = const()[name = string("op_7533_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_687_cast_fp16, y = var_7533_to_fp16)[name = string("aw_chunk_687_cast_fp16")]; fp16 var_7535_to_fp16 = const()[name = string("op_7535_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_689_cast_fp16, y = var_7535_to_fp16)[name = string("aw_chunk_689_cast_fp16")]; fp16 var_7537_to_fp16 = const()[name = string("op_7537_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_691_cast_fp16, y = var_7537_to_fp16)[name = string("aw_chunk_691_cast_fp16")]; fp16 var_7539_to_fp16 = const()[name = string("op_7539_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_693_cast_fp16, y = var_7539_to_fp16)[name = string("aw_chunk_693_cast_fp16")]; fp16 var_7541_to_fp16 = const()[name = string("op_7541_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_695_cast_fp16, y = var_7541_to_fp16)[name = string("aw_chunk_695_cast_fp16")]; fp16 var_7543_to_fp16 = const()[name = string("op_7543_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_697_cast_fp16, y = var_7543_to_fp16)[name = string("aw_chunk_697_cast_fp16")]; fp16 var_7545_to_fp16 = const()[name = string("op_7545_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_699_cast_fp16, y = var_7545_to_fp16)[name = string("aw_chunk_699_cast_fp16")]; fp16 var_7547_to_fp16 = const()[name = string("op_7547_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_701_cast_fp16, y = var_7547_to_fp16)[name = string("aw_chunk_701_cast_fp16")]; fp16 var_7549_to_fp16 = const()[name = string("op_7549_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_703_cast_fp16, y = var_7549_to_fp16)[name = string("aw_chunk_703_cast_fp16")]; fp16 var_7551_to_fp16 = const()[name = string("op_7551_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_705_cast_fp16, y = var_7551_to_fp16)[name = string("aw_chunk_705_cast_fp16")]; fp16 var_7553_to_fp16 = const()[name = string("op_7553_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_707_cast_fp16, y = var_7553_to_fp16)[name = string("aw_chunk_707_cast_fp16")]; fp16 var_7555_to_fp16 = const()[name = string("op_7555_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_709_cast_fp16, y = var_7555_to_fp16)[name = string("aw_chunk_709_cast_fp16")]; fp16 var_7557_to_fp16 = const()[name = string("op_7557_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_711_cast_fp16, y = var_7557_to_fp16)[name = string("aw_chunk_711_cast_fp16")]; fp16 var_7559_to_fp16 = const()[name = string("op_7559_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_713_cast_fp16, y = var_7559_to_fp16)[name = string("aw_chunk_713_cast_fp16")]; fp16 var_7561_to_fp16 = const()[name = string("op_7561_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_715_cast_fp16, y = var_7561_to_fp16)[name = string("aw_chunk_715_cast_fp16")]; fp16 var_7563_to_fp16 = const()[name = string("op_7563_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_717_cast_fp16, y = var_7563_to_fp16)[name = string("aw_chunk_717_cast_fp16")]; fp16 var_7565_to_fp16 = const()[name = string("op_7565_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_719_cast_fp16, y = var_7565_to_fp16)[name = string("aw_chunk_719_cast_fp16")]; fp16 var_7567_to_fp16 = const()[name = string("op_7567_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_721_cast_fp16, y = var_7567_to_fp16)[name = string("aw_chunk_721_cast_fp16")]; fp16 var_7569_to_fp16 = const()[name = string("op_7569_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_723_cast_fp16, y = var_7569_to_fp16)[name = string("aw_chunk_723_cast_fp16")]; fp16 var_7571_to_fp16 = const()[name = string("op_7571_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_725_cast_fp16, y = var_7571_to_fp16)[name = string("aw_chunk_725_cast_fp16")]; fp16 var_7573_to_fp16 = const()[name = string("op_7573_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_727_cast_fp16, y = var_7573_to_fp16)[name = string("aw_chunk_727_cast_fp16")]; fp16 var_7575_to_fp16 = const()[name = string("op_7575_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_729_cast_fp16, y = var_7575_to_fp16)[name = string("aw_chunk_729_cast_fp16")]; fp16 var_7577_to_fp16 = const()[name = string("op_7577_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_731_cast_fp16, y = var_7577_to_fp16)[name = string("aw_chunk_731_cast_fp16")]; fp16 var_7579_to_fp16 = const()[name = string("op_7579_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_733_cast_fp16, y = var_7579_to_fp16)[name = string("aw_chunk_733_cast_fp16")]; fp16 var_7581_to_fp16 = const()[name = string("op_7581_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_735_cast_fp16, y = var_7581_to_fp16)[name = string("aw_chunk_735_cast_fp16")]; fp16 var_7583_to_fp16 = const()[name = string("op_7583_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_737_cast_fp16, y = var_7583_to_fp16)[name = string("aw_chunk_737_cast_fp16")]; fp16 var_7585_to_fp16 = const()[name = string("op_7585_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_739_cast_fp16, y = var_7585_to_fp16)[name = string("aw_chunk_739_cast_fp16")]; fp16 var_7587_to_fp16 = const()[name = string("op_7587_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_741_cast_fp16, y = var_7587_to_fp16)[name = string("aw_chunk_741_cast_fp16")]; fp16 var_7589_to_fp16 = const()[name = string("op_7589_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_743_cast_fp16, y = var_7589_to_fp16)[name = string("aw_chunk_743_cast_fp16")]; fp16 var_7591_to_fp16 = const()[name = string("op_7591_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_745_cast_fp16, y = var_7591_to_fp16)[name = string("aw_chunk_745_cast_fp16")]; fp16 var_7593_to_fp16 = const()[name = string("op_7593_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_747_cast_fp16, y = var_7593_to_fp16)[name = string("aw_chunk_747_cast_fp16")]; fp16 var_7595_to_fp16 = const()[name = string("op_7595_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_749_cast_fp16, y = var_7595_to_fp16)[name = string("aw_chunk_749_cast_fp16")]; fp16 var_7597_to_fp16 = const()[name = string("op_7597_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_751_cast_fp16, y = var_7597_to_fp16)[name = string("aw_chunk_751_cast_fp16")]; fp16 var_7599_to_fp16 = const()[name = string("op_7599_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_753_cast_fp16, y = var_7599_to_fp16)[name = string("aw_chunk_753_cast_fp16")]; fp16 var_7601_to_fp16 = const()[name = string("op_7601_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_755_cast_fp16, y = var_7601_to_fp16)[name = string("aw_chunk_755_cast_fp16")]; fp16 var_7603_to_fp16 = const()[name = string("op_7603_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_757_cast_fp16, y = var_7603_to_fp16)[name = string("aw_chunk_757_cast_fp16")]; fp16 var_7605_to_fp16 = const()[name = string("op_7605_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_759_cast_fp16, y = var_7605_to_fp16)[name = string("aw_chunk_759_cast_fp16")]; fp16 var_7607_to_fp16 = const()[name = string("op_7607_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_761_cast_fp16, y = var_7607_to_fp16)[name = string("aw_chunk_761_cast_fp16")]; fp16 var_7609_to_fp16 = const()[name = string("op_7609_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_763_cast_fp16, y = var_7609_to_fp16)[name = string("aw_chunk_763_cast_fp16")]; fp16 var_7611_to_fp16 = const()[name = string("op_7611_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_765_cast_fp16, y = var_7611_to_fp16)[name = string("aw_chunk_765_cast_fp16")]; fp16 var_7613_to_fp16 = const()[name = string("op_7613_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_767_cast_fp16, y = var_7613_to_fp16)[name = string("aw_chunk_767_cast_fp16")]; tensor var_7615_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_673_cast_fp16)[name = string("op_7615_cast_fp16")]; tensor var_7616_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_675_cast_fp16)[name = string("op_7616_cast_fp16")]; tensor var_7617_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_677_cast_fp16)[name = string("op_7617_cast_fp16")]; tensor var_7618_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_679_cast_fp16)[name = string("op_7618_cast_fp16")]; tensor var_7619_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_681_cast_fp16)[name = string("op_7619_cast_fp16")]; tensor var_7620_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_683_cast_fp16)[name = string("op_7620_cast_fp16")]; tensor var_7621_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_685_cast_fp16)[name = string("op_7621_cast_fp16")]; tensor var_7622_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_687_cast_fp16)[name = string("op_7622_cast_fp16")]; tensor var_7623_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_689_cast_fp16)[name = string("op_7623_cast_fp16")]; tensor var_7624_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_691_cast_fp16)[name = string("op_7624_cast_fp16")]; tensor var_7625_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_693_cast_fp16)[name = string("op_7625_cast_fp16")]; tensor var_7626_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_695_cast_fp16)[name = string("op_7626_cast_fp16")]; tensor var_7627_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_697_cast_fp16)[name = string("op_7627_cast_fp16")]; tensor var_7628_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_699_cast_fp16)[name = string("op_7628_cast_fp16")]; tensor var_7629_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_701_cast_fp16)[name = string("op_7629_cast_fp16")]; tensor var_7630_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_703_cast_fp16)[name = string("op_7630_cast_fp16")]; tensor var_7631_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_705_cast_fp16)[name = string("op_7631_cast_fp16")]; tensor var_7632_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_707_cast_fp16)[name = string("op_7632_cast_fp16")]; tensor var_7633_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_709_cast_fp16)[name = string("op_7633_cast_fp16")]; tensor var_7634_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_711_cast_fp16)[name = string("op_7634_cast_fp16")]; tensor var_7635_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_713_cast_fp16)[name = string("op_7635_cast_fp16")]; tensor var_7636_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_715_cast_fp16)[name = string("op_7636_cast_fp16")]; tensor var_7637_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_717_cast_fp16)[name = string("op_7637_cast_fp16")]; tensor var_7638_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_719_cast_fp16)[name = string("op_7638_cast_fp16")]; tensor var_7639_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_721_cast_fp16)[name = string("op_7639_cast_fp16")]; tensor var_7640_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_723_cast_fp16)[name = string("op_7640_cast_fp16")]; tensor var_7641_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_725_cast_fp16)[name = string("op_7641_cast_fp16")]; tensor var_7642_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_727_cast_fp16)[name = string("op_7642_cast_fp16")]; tensor var_7643_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_729_cast_fp16)[name = string("op_7643_cast_fp16")]; tensor var_7644_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_731_cast_fp16)[name = string("op_7644_cast_fp16")]; tensor var_7645_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_733_cast_fp16)[name = string("op_7645_cast_fp16")]; tensor var_7646_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_735_cast_fp16)[name = string("op_7646_cast_fp16")]; tensor var_7647_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_737_cast_fp16)[name = string("op_7647_cast_fp16")]; tensor var_7648_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_739_cast_fp16)[name = string("op_7648_cast_fp16")]; tensor var_7649_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_741_cast_fp16)[name = string("op_7649_cast_fp16")]; tensor var_7650_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_743_cast_fp16)[name = string("op_7650_cast_fp16")]; tensor var_7651_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_745_cast_fp16)[name = string("op_7651_cast_fp16")]; tensor var_7652_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_747_cast_fp16)[name = string("op_7652_cast_fp16")]; tensor var_7653_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_749_cast_fp16)[name = string("op_7653_cast_fp16")]; tensor var_7654_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_751_cast_fp16)[name = string("op_7654_cast_fp16")]; tensor var_7655_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_753_cast_fp16)[name = string("op_7655_cast_fp16")]; tensor var_7656_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_755_cast_fp16)[name = string("op_7656_cast_fp16")]; tensor var_7657_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_757_cast_fp16)[name = string("op_7657_cast_fp16")]; tensor var_7658_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_759_cast_fp16)[name = string("op_7658_cast_fp16")]; tensor var_7659_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_761_cast_fp16)[name = string("op_7659_cast_fp16")]; tensor var_7660_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_763_cast_fp16)[name = string("op_7660_cast_fp16")]; tensor var_7661_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_765_cast_fp16)[name = string("op_7661_cast_fp16")]; tensor var_7662_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_767_cast_fp16)[name = string("op_7662_cast_fp16")]; string var_7664_equation_0 = const()[name = string("op_7664_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7376_cast_fp16, var_7615_cast_fp16))[name = string("op_7664_cast_fp16")]; string var_7666_equation_0 = const()[name = string("op_7666_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7376_cast_fp16, var_7616_cast_fp16))[name = string("op_7666_cast_fp16")]; string var_7668_equation_0 = const()[name = string("op_7668_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7376_cast_fp16, var_7617_cast_fp16))[name = string("op_7668_cast_fp16")]; string var_7670_equation_0 = const()[name = string("op_7670_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7376_cast_fp16, var_7618_cast_fp16))[name = string("op_7670_cast_fp16")]; string var_7672_equation_0 = const()[name = string("op_7672_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7380_cast_fp16, var_7619_cast_fp16))[name = string("op_7672_cast_fp16")]; string var_7674_equation_0 = const()[name = string("op_7674_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7380_cast_fp16, var_7620_cast_fp16))[name = string("op_7674_cast_fp16")]; string var_7676_equation_0 = const()[name = string("op_7676_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7380_cast_fp16, var_7621_cast_fp16))[name = string("op_7676_cast_fp16")]; string var_7678_equation_0 = const()[name = string("op_7678_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7380_cast_fp16, var_7622_cast_fp16))[name = string("op_7678_cast_fp16")]; string var_7680_equation_0 = const()[name = string("op_7680_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7384_cast_fp16, var_7623_cast_fp16))[name = string("op_7680_cast_fp16")]; string var_7682_equation_0 = const()[name = string("op_7682_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7384_cast_fp16, var_7624_cast_fp16))[name = string("op_7682_cast_fp16")]; string var_7684_equation_0 = const()[name = string("op_7684_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7384_cast_fp16, var_7625_cast_fp16))[name = string("op_7684_cast_fp16")]; string var_7686_equation_0 = const()[name = string("op_7686_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7384_cast_fp16, var_7626_cast_fp16))[name = string("op_7686_cast_fp16")]; string var_7688_equation_0 = const()[name = string("op_7688_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7388_cast_fp16, var_7627_cast_fp16))[name = string("op_7688_cast_fp16")]; string var_7690_equation_0 = const()[name = string("op_7690_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7388_cast_fp16, var_7628_cast_fp16))[name = string("op_7690_cast_fp16")]; string var_7692_equation_0 = const()[name = string("op_7692_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7388_cast_fp16, var_7629_cast_fp16))[name = string("op_7692_cast_fp16")]; string var_7694_equation_0 = const()[name = string("op_7694_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7388_cast_fp16, var_7630_cast_fp16))[name = string("op_7694_cast_fp16")]; string var_7696_equation_0 = const()[name = string("op_7696_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7392_cast_fp16, var_7631_cast_fp16))[name = string("op_7696_cast_fp16")]; string var_7698_equation_0 = const()[name = string("op_7698_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7392_cast_fp16, var_7632_cast_fp16))[name = string("op_7698_cast_fp16")]; string var_7700_equation_0 = const()[name = string("op_7700_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7392_cast_fp16, var_7633_cast_fp16))[name = string("op_7700_cast_fp16")]; string var_7702_equation_0 = const()[name = string("op_7702_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7392_cast_fp16, var_7634_cast_fp16))[name = string("op_7702_cast_fp16")]; string var_7704_equation_0 = const()[name = string("op_7704_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7704_cast_fp16 = einsum(equation = var_7704_equation_0, values = (var_7396_cast_fp16, var_7635_cast_fp16))[name = string("op_7704_cast_fp16")]; string var_7706_equation_0 = const()[name = string("op_7706_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7706_cast_fp16 = einsum(equation = var_7706_equation_0, values = (var_7396_cast_fp16, var_7636_cast_fp16))[name = string("op_7706_cast_fp16")]; string var_7708_equation_0 = const()[name = string("op_7708_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7708_cast_fp16 = einsum(equation = var_7708_equation_0, values = (var_7396_cast_fp16, var_7637_cast_fp16))[name = string("op_7708_cast_fp16")]; string var_7710_equation_0 = const()[name = string("op_7710_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7710_cast_fp16 = einsum(equation = var_7710_equation_0, values = (var_7396_cast_fp16, var_7638_cast_fp16))[name = string("op_7710_cast_fp16")]; string var_7712_equation_0 = const()[name = string("op_7712_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7712_cast_fp16 = einsum(equation = var_7712_equation_0, values = (var_7400_cast_fp16, var_7639_cast_fp16))[name = string("op_7712_cast_fp16")]; string var_7714_equation_0 = const()[name = string("op_7714_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7714_cast_fp16 = einsum(equation = var_7714_equation_0, values = (var_7400_cast_fp16, var_7640_cast_fp16))[name = string("op_7714_cast_fp16")]; string var_7716_equation_0 = const()[name = string("op_7716_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7716_cast_fp16 = einsum(equation = var_7716_equation_0, values = (var_7400_cast_fp16, var_7641_cast_fp16))[name = string("op_7716_cast_fp16")]; string var_7718_equation_0 = const()[name = string("op_7718_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7718_cast_fp16 = einsum(equation = var_7718_equation_0, values = (var_7400_cast_fp16, var_7642_cast_fp16))[name = string("op_7718_cast_fp16")]; string var_7720_equation_0 = const()[name = string("op_7720_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7720_cast_fp16 = einsum(equation = var_7720_equation_0, values = (var_7404_cast_fp16, var_7643_cast_fp16))[name = string("op_7720_cast_fp16")]; string var_7722_equation_0 = const()[name = string("op_7722_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7722_cast_fp16 = einsum(equation = var_7722_equation_0, values = (var_7404_cast_fp16, var_7644_cast_fp16))[name = string("op_7722_cast_fp16")]; string var_7724_equation_0 = const()[name = string("op_7724_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7724_cast_fp16 = einsum(equation = var_7724_equation_0, values = (var_7404_cast_fp16, var_7645_cast_fp16))[name = string("op_7724_cast_fp16")]; string var_7726_equation_0 = const()[name = string("op_7726_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7726_cast_fp16 = einsum(equation = var_7726_equation_0, values = (var_7404_cast_fp16, var_7646_cast_fp16))[name = string("op_7726_cast_fp16")]; string var_7728_equation_0 = const()[name = string("op_7728_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7728_cast_fp16 = einsum(equation = var_7728_equation_0, values = (var_7408_cast_fp16, var_7647_cast_fp16))[name = string("op_7728_cast_fp16")]; string var_7730_equation_0 = const()[name = string("op_7730_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7730_cast_fp16 = einsum(equation = var_7730_equation_0, values = (var_7408_cast_fp16, var_7648_cast_fp16))[name = string("op_7730_cast_fp16")]; string var_7732_equation_0 = const()[name = string("op_7732_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7732_cast_fp16 = einsum(equation = var_7732_equation_0, values = (var_7408_cast_fp16, var_7649_cast_fp16))[name = string("op_7732_cast_fp16")]; string var_7734_equation_0 = const()[name = string("op_7734_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7734_cast_fp16 = einsum(equation = var_7734_equation_0, values = (var_7408_cast_fp16, var_7650_cast_fp16))[name = string("op_7734_cast_fp16")]; string var_7736_equation_0 = const()[name = string("op_7736_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7736_cast_fp16 = einsum(equation = var_7736_equation_0, values = (var_7412_cast_fp16, var_7651_cast_fp16))[name = string("op_7736_cast_fp16")]; string var_7738_equation_0 = const()[name = string("op_7738_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7738_cast_fp16 = einsum(equation = var_7738_equation_0, values = (var_7412_cast_fp16, var_7652_cast_fp16))[name = string("op_7738_cast_fp16")]; string var_7740_equation_0 = const()[name = string("op_7740_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7740_cast_fp16 = einsum(equation = var_7740_equation_0, values = (var_7412_cast_fp16, var_7653_cast_fp16))[name = string("op_7740_cast_fp16")]; string var_7742_equation_0 = const()[name = string("op_7742_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7742_cast_fp16 = einsum(equation = var_7742_equation_0, values = (var_7412_cast_fp16, var_7654_cast_fp16))[name = string("op_7742_cast_fp16")]; string var_7744_equation_0 = const()[name = string("op_7744_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7744_cast_fp16 = einsum(equation = var_7744_equation_0, values = (var_7416_cast_fp16, var_7655_cast_fp16))[name = string("op_7744_cast_fp16")]; string var_7746_equation_0 = const()[name = string("op_7746_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7746_cast_fp16 = einsum(equation = var_7746_equation_0, values = (var_7416_cast_fp16, var_7656_cast_fp16))[name = string("op_7746_cast_fp16")]; string var_7748_equation_0 = const()[name = string("op_7748_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7748_cast_fp16 = einsum(equation = var_7748_equation_0, values = (var_7416_cast_fp16, var_7657_cast_fp16))[name = string("op_7748_cast_fp16")]; string var_7750_equation_0 = const()[name = string("op_7750_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7750_cast_fp16 = einsum(equation = var_7750_equation_0, values = (var_7416_cast_fp16, var_7658_cast_fp16))[name = string("op_7750_cast_fp16")]; string var_7752_equation_0 = const()[name = string("op_7752_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7752_cast_fp16 = einsum(equation = var_7752_equation_0, values = (var_7420_cast_fp16, var_7659_cast_fp16))[name = string("op_7752_cast_fp16")]; string var_7754_equation_0 = const()[name = string("op_7754_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7754_cast_fp16 = einsum(equation = var_7754_equation_0, values = (var_7420_cast_fp16, var_7660_cast_fp16))[name = string("op_7754_cast_fp16")]; string var_7756_equation_0 = const()[name = string("op_7756_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7756_cast_fp16 = einsum(equation = var_7756_equation_0, values = (var_7420_cast_fp16, var_7661_cast_fp16))[name = string("op_7756_cast_fp16")]; string var_7758_equation_0 = const()[name = string("op_7758_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_7758_cast_fp16 = einsum(equation = var_7758_equation_0, values = (var_7420_cast_fp16, var_7662_cast_fp16))[name = string("op_7758_cast_fp16")]; bool var_7760_interleave_0 = const()[name = string("op_7760_interleave_0"), val = bool(false)]; tensor var_7760_cast_fp16 = concat(axis = var_6871, interleave = var_7760_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16))[name = string("op_7760_cast_fp16")]; bool var_7762_interleave_0 = const()[name = string("op_7762_interleave_0"), val = bool(false)]; tensor var_7762_cast_fp16 = concat(axis = var_6871, interleave = var_7762_interleave_0, values = (var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16))[name = string("op_7762_cast_fp16")]; bool var_7764_interleave_0 = const()[name = string("op_7764_interleave_0"), val = bool(false)]; tensor var_7764_cast_fp16 = concat(axis = var_6871, interleave = var_7764_interleave_0, values = (var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16))[name = string("op_7764_cast_fp16")]; bool var_7766_interleave_0 = const()[name = string("op_7766_interleave_0"), val = bool(false)]; tensor var_7766_cast_fp16 = concat(axis = var_6871, interleave = var_7766_interleave_0, values = (var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16))[name = string("op_7766_cast_fp16")]; bool var_7768_interleave_0 = const()[name = string("op_7768_interleave_0"), val = bool(false)]; tensor var_7768_cast_fp16 = concat(axis = var_6871, interleave = var_7768_interleave_0, values = (var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = string("op_7768_cast_fp16")]; bool var_7770_interleave_0 = const()[name = string("op_7770_interleave_0"), val = bool(false)]; tensor var_7770_cast_fp16 = concat(axis = var_6871, interleave = var_7770_interleave_0, values = (var_7704_cast_fp16, var_7706_cast_fp16, var_7708_cast_fp16, var_7710_cast_fp16))[name = string("op_7770_cast_fp16")]; bool var_7772_interleave_0 = const()[name = string("op_7772_interleave_0"), val = bool(false)]; tensor var_7772_cast_fp16 = concat(axis = var_6871, interleave = var_7772_interleave_0, values = (var_7712_cast_fp16, var_7714_cast_fp16, var_7716_cast_fp16, var_7718_cast_fp16))[name = string("op_7772_cast_fp16")]; bool var_7774_interleave_0 = const()[name = string("op_7774_interleave_0"), val = bool(false)]; tensor var_7774_cast_fp16 = concat(axis = var_6871, interleave = var_7774_interleave_0, values = (var_7720_cast_fp16, var_7722_cast_fp16, var_7724_cast_fp16, var_7726_cast_fp16))[name = string("op_7774_cast_fp16")]; bool var_7776_interleave_0 = const()[name = string("op_7776_interleave_0"), val = bool(false)]; tensor var_7776_cast_fp16 = concat(axis = var_6871, interleave = var_7776_interleave_0, values = (var_7728_cast_fp16, var_7730_cast_fp16, var_7732_cast_fp16, var_7734_cast_fp16))[name = string("op_7776_cast_fp16")]; bool var_7778_interleave_0 = const()[name = string("op_7778_interleave_0"), val = bool(false)]; tensor var_7778_cast_fp16 = concat(axis = var_6871, interleave = var_7778_interleave_0, values = (var_7736_cast_fp16, var_7738_cast_fp16, var_7740_cast_fp16, var_7742_cast_fp16))[name = string("op_7778_cast_fp16")]; bool var_7780_interleave_0 = const()[name = string("op_7780_interleave_0"), val = bool(false)]; tensor var_7780_cast_fp16 = concat(axis = var_6871, interleave = var_7780_interleave_0, values = (var_7744_cast_fp16, var_7746_cast_fp16, var_7748_cast_fp16, var_7750_cast_fp16))[name = string("op_7780_cast_fp16")]; bool var_7782_interleave_0 = const()[name = string("op_7782_interleave_0"), val = bool(false)]; tensor var_7782_cast_fp16 = concat(axis = var_6871, interleave = var_7782_interleave_0, values = (var_7752_cast_fp16, var_7754_cast_fp16, var_7756_cast_fp16, var_7758_cast_fp16))[name = string("op_7782_cast_fp16")]; bool input_57_interleave_0 = const()[name = string("input_57_interleave_0"), val = bool(false)]; tensor input_57_cast_fp16 = concat(axis = var_6888, interleave = input_57_interleave_0, values = (var_7760_cast_fp16, var_7762_cast_fp16, var_7764_cast_fp16, var_7766_cast_fp16, var_7768_cast_fp16, var_7770_cast_fp16, var_7772_cast_fp16, var_7774_cast_fp16, var_7776_cast_fp16, var_7778_cast_fp16, var_7780_cast_fp16, var_7782_cast_fp16))[name = string("input_57_cast_fp16")]; string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108989952)))]; tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110169664)))]; tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; fp16 var_7801_to_fp16 = const()[name = string("op_7801_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_7801_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110171264)))]; tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110172864)))]; fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110174464)))]; tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114893120)))]; tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114899328)))]; tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119617984)))]; tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; int32 var_7830 = const()[name = string("op_7830"), val = int32(3)]; int32 var_7847 = const()[name = string("op_7847"), val = int32(1)]; tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; fp16 var_7864_to_fp16 = const()[name = string("op_7864_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_7864_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119619584)))]; tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119621184)))]; fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119622784)))]; tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120802496)))]; tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120804096)))]; tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121983808)))]; tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123163520)))]; tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; tensor var_7902_begin_0 = const()[name = string("op_7902_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7902_end_0 = const()[name = string("op_7902_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7902_end_mask_0 = const()[name = string("op_7902_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = query_17_cast_fp16)[name = string("op_7902_cast_fp16")]; tensor var_7906_begin_0 = const()[name = string("op_7906_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7906_end_0 = const()[name = string("op_7906_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_7906_end_mask_0 = const()[name = string("op_7906_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = query_17_cast_fp16)[name = string("op_7906_cast_fp16")]; tensor var_7910_begin_0 = const()[name = string("op_7910_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7910_end_0 = const()[name = string("op_7910_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_7910_end_mask_0 = const()[name = string("op_7910_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = query_17_cast_fp16)[name = string("op_7910_cast_fp16")]; tensor var_7914_begin_0 = const()[name = string("op_7914_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7914_end_0 = const()[name = string("op_7914_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_7914_end_mask_0 = const()[name = string("op_7914_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = query_17_cast_fp16)[name = string("op_7914_cast_fp16")]; tensor var_7918_begin_0 = const()[name = string("op_7918_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7918_end_0 = const()[name = string("op_7918_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_7918_end_mask_0 = const()[name = string("op_7918_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7918_cast_fp16 = slice_by_index(begin = var_7918_begin_0, end = var_7918_end_0, end_mask = var_7918_end_mask_0, x = query_17_cast_fp16)[name = string("op_7918_cast_fp16")]; tensor var_7922_begin_0 = const()[name = string("op_7922_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7922_end_0 = const()[name = string("op_7922_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_7922_end_mask_0 = const()[name = string("op_7922_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7922_cast_fp16 = slice_by_index(begin = var_7922_begin_0, end = var_7922_end_0, end_mask = var_7922_end_mask_0, x = query_17_cast_fp16)[name = string("op_7922_cast_fp16")]; tensor var_7926_begin_0 = const()[name = string("op_7926_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7926_end_0 = const()[name = string("op_7926_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_7926_end_mask_0 = const()[name = string("op_7926_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7926_cast_fp16 = slice_by_index(begin = var_7926_begin_0, end = var_7926_end_0, end_mask = var_7926_end_mask_0, x = query_17_cast_fp16)[name = string("op_7926_cast_fp16")]; tensor var_7930_begin_0 = const()[name = string("op_7930_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7930_end_0 = const()[name = string("op_7930_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_7930_end_mask_0 = const()[name = string("op_7930_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7930_cast_fp16 = slice_by_index(begin = var_7930_begin_0, end = var_7930_end_0, end_mask = var_7930_end_mask_0, x = query_17_cast_fp16)[name = string("op_7930_cast_fp16")]; tensor var_7934_begin_0 = const()[name = string("op_7934_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7934_end_0 = const()[name = string("op_7934_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_7934_end_mask_0 = const()[name = string("op_7934_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7934_cast_fp16 = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = query_17_cast_fp16)[name = string("op_7934_cast_fp16")]; tensor var_7938_begin_0 = const()[name = string("op_7938_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7938_end_0 = const()[name = string("op_7938_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_7938_end_mask_0 = const()[name = string("op_7938_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7938_cast_fp16 = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = query_17_cast_fp16)[name = string("op_7938_cast_fp16")]; tensor var_7942_begin_0 = const()[name = string("op_7942_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7942_end_0 = const()[name = string("op_7942_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_7942_end_mask_0 = const()[name = string("op_7942_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7942_cast_fp16 = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = query_17_cast_fp16)[name = string("op_7942_cast_fp16")]; tensor var_7946_begin_0 = const()[name = string("op_7946_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7946_end_0 = const()[name = string("op_7946_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_7946_end_mask_0 = const()[name = string("op_7946_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7946_cast_fp16 = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = query_17_cast_fp16)[name = string("op_7946_cast_fp16")]; tensor var_7955_begin_0 = const()[name = string("op_7955_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7955_end_0 = const()[name = string("op_7955_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7955_end_mask_0 = const()[name = string("op_7955_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7955_cast_fp16 = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7955_cast_fp16")]; tensor var_7962_begin_0 = const()[name = string("op_7962_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7962_end_0 = const()[name = string("op_7962_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7962_end_mask_0 = const()[name = string("op_7962_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7962_cast_fp16 = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7962_cast_fp16")]; tensor var_7969_begin_0 = const()[name = string("op_7969_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7969_end_0 = const()[name = string("op_7969_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7969_end_mask_0 = const()[name = string("op_7969_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7969_cast_fp16")]; tensor var_7976_begin_0 = const()[name = string("op_7976_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7976_end_0 = const()[name = string("op_7976_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7976_end_mask_0 = const()[name = string("op_7976_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7976_cast_fp16 = slice_by_index(begin = var_7976_begin_0, end = var_7976_end_0, end_mask = var_7976_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7976_cast_fp16")]; tensor var_7983_begin_0 = const()[name = string("op_7983_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7983_end_0 = const()[name = string("op_7983_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7983_end_mask_0 = const()[name = string("op_7983_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7983_cast_fp16 = slice_by_index(begin = var_7983_begin_0, end = var_7983_end_0, end_mask = var_7983_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7983_cast_fp16")]; tensor var_7990_begin_0 = const()[name = string("op_7990_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7990_end_0 = const()[name = string("op_7990_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7990_end_mask_0 = const()[name = string("op_7990_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7990_cast_fp16 = slice_by_index(begin = var_7990_begin_0, end = var_7990_end_0, end_mask = var_7990_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7990_cast_fp16")]; tensor var_7997_begin_0 = const()[name = string("op_7997_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7997_end_0 = const()[name = string("op_7997_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7997_end_mask_0 = const()[name = string("op_7997_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7997_cast_fp16 = slice_by_index(begin = var_7997_begin_0, end = var_7997_end_0, end_mask = var_7997_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7997_cast_fp16")]; tensor var_8004_begin_0 = const()[name = string("op_8004_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8004_end_0 = const()[name = string("op_8004_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8004_end_mask_0 = const()[name = string("op_8004_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8004_cast_fp16 = slice_by_index(begin = var_8004_begin_0, end = var_8004_end_0, end_mask = var_8004_end_mask_0, x = var_7906_cast_fp16)[name = string("op_8004_cast_fp16")]; tensor var_8011_begin_0 = const()[name = string("op_8011_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8011_end_0 = const()[name = string("op_8011_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8011_end_mask_0 = const()[name = string("op_8011_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8011_cast_fp16 = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8011_cast_fp16")]; tensor var_8018_begin_0 = const()[name = string("op_8018_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8018_end_0 = const()[name = string("op_8018_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8018_end_mask_0 = const()[name = string("op_8018_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8018_cast_fp16 = slice_by_index(begin = var_8018_begin_0, end = var_8018_end_0, end_mask = var_8018_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8018_cast_fp16")]; tensor var_8025_begin_0 = const()[name = string("op_8025_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8025_end_0 = const()[name = string("op_8025_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8025_end_mask_0 = const()[name = string("op_8025_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8025_cast_fp16 = slice_by_index(begin = var_8025_begin_0, end = var_8025_end_0, end_mask = var_8025_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8025_cast_fp16")]; tensor var_8032_begin_0 = const()[name = string("op_8032_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8032_end_0 = const()[name = string("op_8032_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8032_end_mask_0 = const()[name = string("op_8032_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8032_cast_fp16 = slice_by_index(begin = var_8032_begin_0, end = var_8032_end_0, end_mask = var_8032_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8032_cast_fp16")]; tensor var_8039_begin_0 = const()[name = string("op_8039_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8039_end_0 = const()[name = string("op_8039_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8039_end_mask_0 = const()[name = string("op_8039_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8039_cast_fp16 = slice_by_index(begin = var_8039_begin_0, end = var_8039_end_0, end_mask = var_8039_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8039_cast_fp16")]; tensor var_8046_begin_0 = const()[name = string("op_8046_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8046_end_0 = const()[name = string("op_8046_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8046_end_mask_0 = const()[name = string("op_8046_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8046_cast_fp16 = slice_by_index(begin = var_8046_begin_0, end = var_8046_end_0, end_mask = var_8046_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8046_cast_fp16")]; tensor var_8053_begin_0 = const()[name = string("op_8053_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8053_end_0 = const()[name = string("op_8053_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8053_end_mask_0 = const()[name = string("op_8053_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8053_cast_fp16 = slice_by_index(begin = var_8053_begin_0, end = var_8053_end_0, end_mask = var_8053_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8053_cast_fp16")]; tensor var_8060_begin_0 = const()[name = string("op_8060_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8060_end_0 = const()[name = string("op_8060_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8060_end_mask_0 = const()[name = string("op_8060_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8060_cast_fp16 = slice_by_index(begin = var_8060_begin_0, end = var_8060_end_0, end_mask = var_8060_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8060_cast_fp16")]; tensor var_8067_begin_0 = const()[name = string("op_8067_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8067_end_0 = const()[name = string("op_8067_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8067_end_mask_0 = const()[name = string("op_8067_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8067_cast_fp16 = slice_by_index(begin = var_8067_begin_0, end = var_8067_end_0, end_mask = var_8067_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8067_cast_fp16")]; tensor var_8074_begin_0 = const()[name = string("op_8074_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8074_end_0 = const()[name = string("op_8074_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8074_end_mask_0 = const()[name = string("op_8074_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8074_cast_fp16 = slice_by_index(begin = var_8074_begin_0, end = var_8074_end_0, end_mask = var_8074_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8074_cast_fp16")]; tensor var_8081_begin_0 = const()[name = string("op_8081_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8081_end_0 = const()[name = string("op_8081_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8081_end_mask_0 = const()[name = string("op_8081_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8081_cast_fp16 = slice_by_index(begin = var_8081_begin_0, end = var_8081_end_0, end_mask = var_8081_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8081_cast_fp16")]; tensor var_8088_begin_0 = const()[name = string("op_8088_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8088_end_0 = const()[name = string("op_8088_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8088_end_mask_0 = const()[name = string("op_8088_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8088_cast_fp16 = slice_by_index(begin = var_8088_begin_0, end = var_8088_end_0, end_mask = var_8088_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8088_cast_fp16")]; tensor var_8095_begin_0 = const()[name = string("op_8095_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8095_end_0 = const()[name = string("op_8095_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8095_end_mask_0 = const()[name = string("op_8095_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8095_cast_fp16 = slice_by_index(begin = var_8095_begin_0, end = var_8095_end_0, end_mask = var_8095_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8095_cast_fp16")]; tensor var_8102_begin_0 = const()[name = string("op_8102_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8102_end_0 = const()[name = string("op_8102_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8102_end_mask_0 = const()[name = string("op_8102_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8102_cast_fp16 = slice_by_index(begin = var_8102_begin_0, end = var_8102_end_0, end_mask = var_8102_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8102_cast_fp16")]; tensor var_8109_begin_0 = const()[name = string("op_8109_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8109_end_0 = const()[name = string("op_8109_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8109_end_mask_0 = const()[name = string("op_8109_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8109_cast_fp16 = slice_by_index(begin = var_8109_begin_0, end = var_8109_end_0, end_mask = var_8109_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8109_cast_fp16")]; tensor var_8116_begin_0 = const()[name = string("op_8116_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8116_end_0 = const()[name = string("op_8116_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8116_end_mask_0 = const()[name = string("op_8116_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8116_cast_fp16 = slice_by_index(begin = var_8116_begin_0, end = var_8116_end_0, end_mask = var_8116_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8116_cast_fp16")]; tensor var_8123_begin_0 = const()[name = string("op_8123_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8123_end_0 = const()[name = string("op_8123_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8123_end_mask_0 = const()[name = string("op_8123_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8123_cast_fp16 = slice_by_index(begin = var_8123_begin_0, end = var_8123_end_0, end_mask = var_8123_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8123_cast_fp16")]; tensor var_8130_begin_0 = const()[name = string("op_8130_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8130_end_0 = const()[name = string("op_8130_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8130_end_mask_0 = const()[name = string("op_8130_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8130_cast_fp16 = slice_by_index(begin = var_8130_begin_0, end = var_8130_end_0, end_mask = var_8130_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8130_cast_fp16")]; tensor var_8137_begin_0 = const()[name = string("op_8137_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8137_end_0 = const()[name = string("op_8137_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8137_end_mask_0 = const()[name = string("op_8137_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8137_cast_fp16 = slice_by_index(begin = var_8137_begin_0, end = var_8137_end_0, end_mask = var_8137_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8137_cast_fp16")]; tensor var_8144_begin_0 = const()[name = string("op_8144_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8144_end_0 = const()[name = string("op_8144_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8144_end_mask_0 = const()[name = string("op_8144_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8144_cast_fp16 = slice_by_index(begin = var_8144_begin_0, end = var_8144_end_0, end_mask = var_8144_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8144_cast_fp16")]; tensor var_8151_begin_0 = const()[name = string("op_8151_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8151_end_0 = const()[name = string("op_8151_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8151_end_mask_0 = const()[name = string("op_8151_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8151_cast_fp16 = slice_by_index(begin = var_8151_begin_0, end = var_8151_end_0, end_mask = var_8151_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8151_cast_fp16")]; tensor var_8158_begin_0 = const()[name = string("op_8158_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8158_end_0 = const()[name = string("op_8158_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8158_end_mask_0 = const()[name = string("op_8158_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8158_cast_fp16 = slice_by_index(begin = var_8158_begin_0, end = var_8158_end_0, end_mask = var_8158_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8158_cast_fp16")]; tensor var_8165_begin_0 = const()[name = string("op_8165_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8165_end_0 = const()[name = string("op_8165_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8165_end_mask_0 = const()[name = string("op_8165_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8165_cast_fp16 = slice_by_index(begin = var_8165_begin_0, end = var_8165_end_0, end_mask = var_8165_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8165_cast_fp16")]; tensor var_8172_begin_0 = const()[name = string("op_8172_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8172_end_0 = const()[name = string("op_8172_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8172_end_mask_0 = const()[name = string("op_8172_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8172_cast_fp16 = slice_by_index(begin = var_8172_begin_0, end = var_8172_end_0, end_mask = var_8172_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8172_cast_fp16")]; tensor var_8179_begin_0 = const()[name = string("op_8179_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8179_end_0 = const()[name = string("op_8179_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8179_end_mask_0 = const()[name = string("op_8179_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8179_cast_fp16 = slice_by_index(begin = var_8179_begin_0, end = var_8179_end_0, end_mask = var_8179_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8179_cast_fp16")]; tensor var_8186_begin_0 = const()[name = string("op_8186_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8186_end_0 = const()[name = string("op_8186_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8186_end_mask_0 = const()[name = string("op_8186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8186_cast_fp16 = slice_by_index(begin = var_8186_begin_0, end = var_8186_end_0, end_mask = var_8186_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8186_cast_fp16")]; tensor var_8193_begin_0 = const()[name = string("op_8193_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8193_end_0 = const()[name = string("op_8193_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8193_end_mask_0 = const()[name = string("op_8193_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8193_cast_fp16 = slice_by_index(begin = var_8193_begin_0, end = var_8193_end_0, end_mask = var_8193_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8193_cast_fp16")]; tensor var_8200_begin_0 = const()[name = string("op_8200_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8200_end_0 = const()[name = string("op_8200_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8200_end_mask_0 = const()[name = string("op_8200_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8200_cast_fp16 = slice_by_index(begin = var_8200_begin_0, end = var_8200_end_0, end_mask = var_8200_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8200_cast_fp16")]; tensor var_8207_begin_0 = const()[name = string("op_8207_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8207_end_0 = const()[name = string("op_8207_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8207_end_mask_0 = const()[name = string("op_8207_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8207_cast_fp16 = slice_by_index(begin = var_8207_begin_0, end = var_8207_end_0, end_mask = var_8207_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8207_cast_fp16")]; tensor var_8214_begin_0 = const()[name = string("op_8214_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8214_end_0 = const()[name = string("op_8214_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8214_end_mask_0 = const()[name = string("op_8214_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8214_cast_fp16 = slice_by_index(begin = var_8214_begin_0, end = var_8214_end_0, end_mask = var_8214_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8214_cast_fp16")]; tensor var_8221_begin_0 = const()[name = string("op_8221_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8221_end_0 = const()[name = string("op_8221_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8221_end_mask_0 = const()[name = string("op_8221_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8221_cast_fp16 = slice_by_index(begin = var_8221_begin_0, end = var_8221_end_0, end_mask = var_8221_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8221_cast_fp16")]; tensor var_8228_begin_0 = const()[name = string("op_8228_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8228_end_0 = const()[name = string("op_8228_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8228_end_mask_0 = const()[name = string("op_8228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8228_cast_fp16 = slice_by_index(begin = var_8228_begin_0, end = var_8228_end_0, end_mask = var_8228_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8228_cast_fp16")]; tensor var_8235_begin_0 = const()[name = string("op_8235_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8235_end_0 = const()[name = string("op_8235_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8235_end_mask_0 = const()[name = string("op_8235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8235_cast_fp16 = slice_by_index(begin = var_8235_begin_0, end = var_8235_end_0, end_mask = var_8235_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8235_cast_fp16")]; tensor var_8242_begin_0 = const()[name = string("op_8242_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8242_end_0 = const()[name = string("op_8242_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8242_end_mask_0 = const()[name = string("op_8242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8242_cast_fp16")]; tensor var_8249_begin_0 = const()[name = string("op_8249_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8249_end_0 = const()[name = string("op_8249_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8249_end_mask_0 = const()[name = string("op_8249_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8249_cast_fp16 = slice_by_index(begin = var_8249_begin_0, end = var_8249_end_0, end_mask = var_8249_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8249_cast_fp16")]; tensor var_8256_begin_0 = const()[name = string("op_8256_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8256_end_0 = const()[name = string("op_8256_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8256_end_mask_0 = const()[name = string("op_8256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8256_cast_fp16 = slice_by_index(begin = var_8256_begin_0, end = var_8256_end_0, end_mask = var_8256_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8256_cast_fp16")]; tensor var_8263_begin_0 = const()[name = string("op_8263_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8263_end_0 = const()[name = string("op_8263_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8263_end_mask_0 = const()[name = string("op_8263_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8263_cast_fp16 = slice_by_index(begin = var_8263_begin_0, end = var_8263_end_0, end_mask = var_8263_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8263_cast_fp16")]; tensor var_8270_begin_0 = const()[name = string("op_8270_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8270_end_0 = const()[name = string("op_8270_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8270_end_mask_0 = const()[name = string("op_8270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8270_cast_fp16")]; tensor var_8277_begin_0 = const()[name = string("op_8277_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8277_end_0 = const()[name = string("op_8277_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8277_end_mask_0 = const()[name = string("op_8277_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8277_cast_fp16 = slice_by_index(begin = var_8277_begin_0, end = var_8277_end_0, end_mask = var_8277_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8277_cast_fp16")]; tensor var_8284_begin_0 = const()[name = string("op_8284_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8284_end_0 = const()[name = string("op_8284_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8284_end_mask_0 = const()[name = string("op_8284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8284_cast_fp16 = slice_by_index(begin = var_8284_begin_0, end = var_8284_end_0, end_mask = var_8284_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8284_cast_fp16")]; tensor k_17_perm_0 = const()[name = string("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8289_begin_0 = const()[name = string("op_8289_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8289_end_0 = const()[name = string("op_8289_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_8289_end_mask_0 = const()[name = string("op_8289_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_17_cast_fp16 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = string("transpose_3")]; tensor var_8289_cast_fp16 = slice_by_index(begin = var_8289_begin_0, end = var_8289_end_0, end_mask = var_8289_end_mask_0, x = k_17_cast_fp16)[name = string("op_8289_cast_fp16")]; tensor var_8293_begin_0 = const()[name = string("op_8293_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_8293_end_0 = const()[name = string("op_8293_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_8293_end_mask_0 = const()[name = string("op_8293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8293_cast_fp16 = slice_by_index(begin = var_8293_begin_0, end = var_8293_end_0, end_mask = var_8293_end_mask_0, x = k_17_cast_fp16)[name = string("op_8293_cast_fp16")]; tensor var_8297_begin_0 = const()[name = string("op_8297_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_8297_end_0 = const()[name = string("op_8297_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_8297_end_mask_0 = const()[name = string("op_8297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8297_cast_fp16 = slice_by_index(begin = var_8297_begin_0, end = var_8297_end_0, end_mask = var_8297_end_mask_0, x = k_17_cast_fp16)[name = string("op_8297_cast_fp16")]; tensor var_8301_begin_0 = const()[name = string("op_8301_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_8301_end_0 = const()[name = string("op_8301_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_8301_end_mask_0 = const()[name = string("op_8301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8301_cast_fp16 = slice_by_index(begin = var_8301_begin_0, end = var_8301_end_0, end_mask = var_8301_end_mask_0, x = k_17_cast_fp16)[name = string("op_8301_cast_fp16")]; tensor var_8305_begin_0 = const()[name = string("op_8305_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_8305_end_0 = const()[name = string("op_8305_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_8305_end_mask_0 = const()[name = string("op_8305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8305_cast_fp16 = slice_by_index(begin = var_8305_begin_0, end = var_8305_end_0, end_mask = var_8305_end_mask_0, x = k_17_cast_fp16)[name = string("op_8305_cast_fp16")]; tensor var_8309_begin_0 = const()[name = string("op_8309_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8309_end_0 = const()[name = string("op_8309_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_8309_end_mask_0 = const()[name = string("op_8309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8309_cast_fp16 = slice_by_index(begin = var_8309_begin_0, end = var_8309_end_0, end_mask = var_8309_end_mask_0, x = k_17_cast_fp16)[name = string("op_8309_cast_fp16")]; tensor var_8313_begin_0 = const()[name = string("op_8313_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_8313_end_0 = const()[name = string("op_8313_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_8313_end_mask_0 = const()[name = string("op_8313_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8313_cast_fp16 = slice_by_index(begin = var_8313_begin_0, end = var_8313_end_0, end_mask = var_8313_end_mask_0, x = k_17_cast_fp16)[name = string("op_8313_cast_fp16")]; tensor var_8317_begin_0 = const()[name = string("op_8317_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_8317_end_0 = const()[name = string("op_8317_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_8317_end_mask_0 = const()[name = string("op_8317_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8317_cast_fp16 = slice_by_index(begin = var_8317_begin_0, end = var_8317_end_0, end_mask = var_8317_end_mask_0, x = k_17_cast_fp16)[name = string("op_8317_cast_fp16")]; tensor var_8321_begin_0 = const()[name = string("op_8321_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8321_end_0 = const()[name = string("op_8321_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_8321_end_mask_0 = const()[name = string("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = k_17_cast_fp16)[name = string("op_8321_cast_fp16")]; tensor var_8325_begin_0 = const()[name = string("op_8325_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_8325_end_0 = const()[name = string("op_8325_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_8325_end_mask_0 = const()[name = string("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = k_17_cast_fp16)[name = string("op_8325_cast_fp16")]; tensor var_8329_begin_0 = const()[name = string("op_8329_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_8329_end_0 = const()[name = string("op_8329_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_8329_end_mask_0 = const()[name = string("op_8329_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = k_17_cast_fp16)[name = string("op_8329_cast_fp16")]; tensor var_8333_begin_0 = const()[name = string("op_8333_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_8333_end_0 = const()[name = string("op_8333_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_8333_end_mask_0 = const()[name = string("op_8333_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = k_17_cast_fp16)[name = string("op_8333_cast_fp16")]; tensor var_8335_begin_0 = const()[name = string("op_8335_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8335_end_0 = const()[name = string("op_8335_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8335_end_mask_0 = const()[name = string("op_8335_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = value_17_cast_fp16)[name = string("op_8335_cast_fp16")]; tensor var_8339_begin_0 = const()[name = string("op_8339_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8339_end_0 = const()[name = string("op_8339_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_8339_end_mask_0 = const()[name = string("op_8339_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = value_17_cast_fp16)[name = string("op_8339_cast_fp16")]; tensor var_8343_begin_0 = const()[name = string("op_8343_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8343_end_0 = const()[name = string("op_8343_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_8343_end_mask_0 = const()[name = string("op_8343_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8343_cast_fp16 = slice_by_index(begin = var_8343_begin_0, end = var_8343_end_0, end_mask = var_8343_end_mask_0, x = value_17_cast_fp16)[name = string("op_8343_cast_fp16")]; tensor var_8347_begin_0 = const()[name = string("op_8347_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8347_end_0 = const()[name = string("op_8347_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_8347_end_mask_0 = const()[name = string("op_8347_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8347_cast_fp16 = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = value_17_cast_fp16)[name = string("op_8347_cast_fp16")]; tensor var_8351_begin_0 = const()[name = string("op_8351_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8351_end_0 = const()[name = string("op_8351_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_8351_end_mask_0 = const()[name = string("op_8351_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8351_cast_fp16 = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = value_17_cast_fp16)[name = string("op_8351_cast_fp16")]; tensor var_8355_begin_0 = const()[name = string("op_8355_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8355_end_0 = const()[name = string("op_8355_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_8355_end_mask_0 = const()[name = string("op_8355_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8355_cast_fp16 = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = value_17_cast_fp16)[name = string("op_8355_cast_fp16")]; tensor var_8359_begin_0 = const()[name = string("op_8359_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8359_end_0 = const()[name = string("op_8359_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_8359_end_mask_0 = const()[name = string("op_8359_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8359_cast_fp16 = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = value_17_cast_fp16)[name = string("op_8359_cast_fp16")]; tensor var_8363_begin_0 = const()[name = string("op_8363_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8363_end_0 = const()[name = string("op_8363_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_8363_end_mask_0 = const()[name = string("op_8363_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8363_cast_fp16 = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = value_17_cast_fp16)[name = string("op_8363_cast_fp16")]; tensor var_8367_begin_0 = const()[name = string("op_8367_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8367_end_0 = const()[name = string("op_8367_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_8367_end_mask_0 = const()[name = string("op_8367_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8367_cast_fp16 = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = value_17_cast_fp16)[name = string("op_8367_cast_fp16")]; tensor var_8371_begin_0 = const()[name = string("op_8371_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8371_end_0 = const()[name = string("op_8371_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_8371_end_mask_0 = const()[name = string("op_8371_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8371_cast_fp16 = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = value_17_cast_fp16)[name = string("op_8371_cast_fp16")]; tensor var_8375_begin_0 = const()[name = string("op_8375_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8375_end_0 = const()[name = string("op_8375_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_8375_end_mask_0 = const()[name = string("op_8375_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8375_cast_fp16 = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = value_17_cast_fp16)[name = string("op_8375_cast_fp16")]; tensor var_8379_begin_0 = const()[name = string("op_8379_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8379_end_0 = const()[name = string("op_8379_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_8379_end_mask_0 = const()[name = string("op_8379_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8379_cast_fp16 = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = value_17_cast_fp16)[name = string("op_8379_cast_fp16")]; string _SplitHeadsQ__mh_w_769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_769_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_769_equation_0, values = (var_8289_cast_fp16, var_7955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_769_cast_fp16")]; string _SplitHeadsQ__mh_w_771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_771_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_771_equation_0, values = (var_8289_cast_fp16, var_7962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_771_cast_fp16")]; string _SplitHeadsQ__mh_w_773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_773_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_773_equation_0, values = (var_8289_cast_fp16, var_7969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_773_cast_fp16")]; string _SplitHeadsQ__mh_w_775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_775_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_775_equation_0, values = (var_8289_cast_fp16, var_7976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_775_cast_fp16")]; string _SplitHeadsQ__mh_w_777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_777_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_777_equation_0, values = (var_8293_cast_fp16, var_7983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_777_cast_fp16")]; string _SplitHeadsQ__mh_w_779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_779_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_779_equation_0, values = (var_8293_cast_fp16, var_7990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_779_cast_fp16")]; string _SplitHeadsQ__mh_w_781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_781_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_781_equation_0, values = (var_8293_cast_fp16, var_7997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_781_cast_fp16")]; string _SplitHeadsQ__mh_w_783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_783_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_783_equation_0, values = (var_8293_cast_fp16, var_8004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_783_cast_fp16")]; string _SplitHeadsQ__mh_w_785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_785_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_785_equation_0, values = (var_8297_cast_fp16, var_8011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_785_cast_fp16")]; string _SplitHeadsQ__mh_w_787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_787_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_787_equation_0, values = (var_8297_cast_fp16, var_8018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_787_cast_fp16")]; string _SplitHeadsQ__mh_w_789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_789_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_789_equation_0, values = (var_8297_cast_fp16, var_8025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_789_cast_fp16")]; string _SplitHeadsQ__mh_w_791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_791_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_791_equation_0, values = (var_8297_cast_fp16, var_8032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_791_cast_fp16")]; string _SplitHeadsQ__mh_w_793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_793_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_793_equation_0, values = (var_8301_cast_fp16, var_8039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_793_cast_fp16")]; string _SplitHeadsQ__mh_w_795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_795_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_795_equation_0, values = (var_8301_cast_fp16, var_8046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_795_cast_fp16")]; string _SplitHeadsQ__mh_w_797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_797_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_797_equation_0, values = (var_8301_cast_fp16, var_8053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_797_cast_fp16")]; string _SplitHeadsQ__mh_w_799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_799_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_799_equation_0, values = (var_8301_cast_fp16, var_8060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_799_cast_fp16")]; string _SplitHeadsQ__mh_w_801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_801_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_801_equation_0, values = (var_8305_cast_fp16, var_8067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_801_cast_fp16")]; string _SplitHeadsQ__mh_w_803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_803_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_803_equation_0, values = (var_8305_cast_fp16, var_8074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_803_cast_fp16")]; string _SplitHeadsQ__mh_w_805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_805_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_805_equation_0, values = (var_8305_cast_fp16, var_8081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_805_cast_fp16")]; string _SplitHeadsQ__mh_w_807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_807_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_807_equation_0, values = (var_8305_cast_fp16, var_8088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_807_cast_fp16")]; string _SplitHeadsQ__mh_w_809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_809_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_809_equation_0, values = (var_8309_cast_fp16, var_8095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_809_cast_fp16")]; string _SplitHeadsQ__mh_w_811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_811_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_811_equation_0, values = (var_8309_cast_fp16, var_8102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_811_cast_fp16")]; string _SplitHeadsQ__mh_w_813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_813_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_813_equation_0, values = (var_8309_cast_fp16, var_8109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_813_cast_fp16")]; string _SplitHeadsQ__mh_w_815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_815_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_815_equation_0, values = (var_8309_cast_fp16, var_8116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_815_cast_fp16")]; string _SplitHeadsQ__mh_w_817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_817_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_817_equation_0, values = (var_8313_cast_fp16, var_8123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_817_cast_fp16")]; string _SplitHeadsQ__mh_w_819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_819_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_819_equation_0, values = (var_8313_cast_fp16, var_8130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_819_cast_fp16")]; string _SplitHeadsQ__mh_w_821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_821_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_821_equation_0, values = (var_8313_cast_fp16, var_8137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_821_cast_fp16")]; string _SplitHeadsQ__mh_w_823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_823_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_823_equation_0, values = (var_8313_cast_fp16, var_8144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_823_cast_fp16")]; string _SplitHeadsQ__mh_w_825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_825_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_825_equation_0, values = (var_8317_cast_fp16, var_8151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_825_cast_fp16")]; string _SplitHeadsQ__mh_w_827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_827_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_827_equation_0, values = (var_8317_cast_fp16, var_8158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_827_cast_fp16")]; string _SplitHeadsQ__mh_w_829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_829_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_829_equation_0, values = (var_8317_cast_fp16, var_8165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_829_cast_fp16")]; string _SplitHeadsQ__mh_w_831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_831_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_831_equation_0, values = (var_8317_cast_fp16, var_8172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_831_cast_fp16")]; string _SplitHeadsQ__mh_w_833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_833_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_833_equation_0, values = (var_8321_cast_fp16, var_8179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_833_cast_fp16")]; string _SplitHeadsQ__mh_w_835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_835_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_835_equation_0, values = (var_8321_cast_fp16, var_8186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_835_cast_fp16")]; string _SplitHeadsQ__mh_w_837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_837_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_837_equation_0, values = (var_8321_cast_fp16, var_8193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_837_cast_fp16")]; string _SplitHeadsQ__mh_w_839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_839_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_839_equation_0, values = (var_8321_cast_fp16, var_8200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_839_cast_fp16")]; string _SplitHeadsQ__mh_w_841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_841_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_841_equation_0, values = (var_8325_cast_fp16, var_8207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_841_cast_fp16")]; string _SplitHeadsQ__mh_w_843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_843_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_843_equation_0, values = (var_8325_cast_fp16, var_8214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_843_cast_fp16")]; string _SplitHeadsQ__mh_w_845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_845_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_845_equation_0, values = (var_8325_cast_fp16, var_8221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_845_cast_fp16")]; string _SplitHeadsQ__mh_w_847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_847_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_847_equation_0, values = (var_8325_cast_fp16, var_8228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_847_cast_fp16")]; string _SplitHeadsQ__mh_w_849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_849_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_849_equation_0, values = (var_8329_cast_fp16, var_8235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_849_cast_fp16")]; string _SplitHeadsQ__mh_w_851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_851_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_851_equation_0, values = (var_8329_cast_fp16, var_8242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_851_cast_fp16")]; string _SplitHeadsQ__mh_w_853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_853_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_853_equation_0, values = (var_8329_cast_fp16, var_8249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_853_cast_fp16")]; string _SplitHeadsQ__mh_w_855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_855_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_855_equation_0, values = (var_8329_cast_fp16, var_8256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_855_cast_fp16")]; string _SplitHeadsQ__mh_w_857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_857_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_857_equation_0, values = (var_8333_cast_fp16, var_8263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_857_cast_fp16")]; string _SplitHeadsQ__mh_w_859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_859_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_859_equation_0, values = (var_8333_cast_fp16, var_8270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_859_cast_fp16")]; string _SplitHeadsQ__mh_w_861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_861_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_861_equation_0, values = (var_8333_cast_fp16, var_8277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_861_cast_fp16")]; string _SplitHeadsQ__mh_w_863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_863_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_863_equation_0, values = (var_8333_cast_fp16, var_8284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_863_cast_fp16")]; fp16 var_8478_to_fp16 = const()[name = string("op_8478_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_769_cast_fp16, y = var_8478_to_fp16)[name = string("aw_chunk_769_cast_fp16")]; fp16 var_8480_to_fp16 = const()[name = string("op_8480_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_771_cast_fp16, y = var_8480_to_fp16)[name = string("aw_chunk_771_cast_fp16")]; fp16 var_8482_to_fp16 = const()[name = string("op_8482_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_773_cast_fp16, y = var_8482_to_fp16)[name = string("aw_chunk_773_cast_fp16")]; fp16 var_8484_to_fp16 = const()[name = string("op_8484_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_775_cast_fp16, y = var_8484_to_fp16)[name = string("aw_chunk_775_cast_fp16")]; fp16 var_8486_to_fp16 = const()[name = string("op_8486_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_777_cast_fp16, y = var_8486_to_fp16)[name = string("aw_chunk_777_cast_fp16")]; fp16 var_8488_to_fp16 = const()[name = string("op_8488_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_779_cast_fp16, y = var_8488_to_fp16)[name = string("aw_chunk_779_cast_fp16")]; fp16 var_8490_to_fp16 = const()[name = string("op_8490_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_781_cast_fp16, y = var_8490_to_fp16)[name = string("aw_chunk_781_cast_fp16")]; fp16 var_8492_to_fp16 = const()[name = string("op_8492_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_783_cast_fp16, y = var_8492_to_fp16)[name = string("aw_chunk_783_cast_fp16")]; fp16 var_8494_to_fp16 = const()[name = string("op_8494_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_785_cast_fp16, y = var_8494_to_fp16)[name = string("aw_chunk_785_cast_fp16")]; fp16 var_8496_to_fp16 = const()[name = string("op_8496_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_787_cast_fp16, y = var_8496_to_fp16)[name = string("aw_chunk_787_cast_fp16")]; fp16 var_8498_to_fp16 = const()[name = string("op_8498_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_789_cast_fp16, y = var_8498_to_fp16)[name = string("aw_chunk_789_cast_fp16")]; fp16 var_8500_to_fp16 = const()[name = string("op_8500_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_791_cast_fp16, y = var_8500_to_fp16)[name = string("aw_chunk_791_cast_fp16")]; fp16 var_8502_to_fp16 = const()[name = string("op_8502_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_793_cast_fp16, y = var_8502_to_fp16)[name = string("aw_chunk_793_cast_fp16")]; fp16 var_8504_to_fp16 = const()[name = string("op_8504_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_795_cast_fp16, y = var_8504_to_fp16)[name = string("aw_chunk_795_cast_fp16")]; fp16 var_8506_to_fp16 = const()[name = string("op_8506_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_797_cast_fp16, y = var_8506_to_fp16)[name = string("aw_chunk_797_cast_fp16")]; fp16 var_8508_to_fp16 = const()[name = string("op_8508_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_799_cast_fp16, y = var_8508_to_fp16)[name = string("aw_chunk_799_cast_fp16")]; fp16 var_8510_to_fp16 = const()[name = string("op_8510_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_801_cast_fp16, y = var_8510_to_fp16)[name = string("aw_chunk_801_cast_fp16")]; fp16 var_8512_to_fp16 = const()[name = string("op_8512_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_803_cast_fp16, y = var_8512_to_fp16)[name = string("aw_chunk_803_cast_fp16")]; fp16 var_8514_to_fp16 = const()[name = string("op_8514_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_805_cast_fp16, y = var_8514_to_fp16)[name = string("aw_chunk_805_cast_fp16")]; fp16 var_8516_to_fp16 = const()[name = string("op_8516_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_807_cast_fp16, y = var_8516_to_fp16)[name = string("aw_chunk_807_cast_fp16")]; fp16 var_8518_to_fp16 = const()[name = string("op_8518_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_809_cast_fp16, y = var_8518_to_fp16)[name = string("aw_chunk_809_cast_fp16")]; fp16 var_8520_to_fp16 = const()[name = string("op_8520_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_811_cast_fp16, y = var_8520_to_fp16)[name = string("aw_chunk_811_cast_fp16")]; fp16 var_8522_to_fp16 = const()[name = string("op_8522_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_813_cast_fp16, y = var_8522_to_fp16)[name = string("aw_chunk_813_cast_fp16")]; fp16 var_8524_to_fp16 = const()[name = string("op_8524_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_815_cast_fp16, y = var_8524_to_fp16)[name = string("aw_chunk_815_cast_fp16")]; fp16 var_8526_to_fp16 = const()[name = string("op_8526_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_817_cast_fp16, y = var_8526_to_fp16)[name = string("aw_chunk_817_cast_fp16")]; fp16 var_8528_to_fp16 = const()[name = string("op_8528_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_819_cast_fp16, y = var_8528_to_fp16)[name = string("aw_chunk_819_cast_fp16")]; fp16 var_8530_to_fp16 = const()[name = string("op_8530_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_821_cast_fp16, y = var_8530_to_fp16)[name = string("aw_chunk_821_cast_fp16")]; fp16 var_8532_to_fp16 = const()[name = string("op_8532_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_823_cast_fp16, y = var_8532_to_fp16)[name = string("aw_chunk_823_cast_fp16")]; fp16 var_8534_to_fp16 = const()[name = string("op_8534_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_825_cast_fp16, y = var_8534_to_fp16)[name = string("aw_chunk_825_cast_fp16")]; fp16 var_8536_to_fp16 = const()[name = string("op_8536_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_827_cast_fp16, y = var_8536_to_fp16)[name = string("aw_chunk_827_cast_fp16")]; fp16 var_8538_to_fp16 = const()[name = string("op_8538_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_829_cast_fp16, y = var_8538_to_fp16)[name = string("aw_chunk_829_cast_fp16")]; fp16 var_8540_to_fp16 = const()[name = string("op_8540_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_831_cast_fp16, y = var_8540_to_fp16)[name = string("aw_chunk_831_cast_fp16")]; fp16 var_8542_to_fp16 = const()[name = string("op_8542_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_833_cast_fp16, y = var_8542_to_fp16)[name = string("aw_chunk_833_cast_fp16")]; fp16 var_8544_to_fp16 = const()[name = string("op_8544_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_835_cast_fp16, y = var_8544_to_fp16)[name = string("aw_chunk_835_cast_fp16")]; fp16 var_8546_to_fp16 = const()[name = string("op_8546_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_837_cast_fp16, y = var_8546_to_fp16)[name = string("aw_chunk_837_cast_fp16")]; fp16 var_8548_to_fp16 = const()[name = string("op_8548_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_839_cast_fp16, y = var_8548_to_fp16)[name = string("aw_chunk_839_cast_fp16")]; fp16 var_8550_to_fp16 = const()[name = string("op_8550_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_841_cast_fp16, y = var_8550_to_fp16)[name = string("aw_chunk_841_cast_fp16")]; fp16 var_8552_to_fp16 = const()[name = string("op_8552_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_843_cast_fp16, y = var_8552_to_fp16)[name = string("aw_chunk_843_cast_fp16")]; fp16 var_8554_to_fp16 = const()[name = string("op_8554_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_845_cast_fp16, y = var_8554_to_fp16)[name = string("aw_chunk_845_cast_fp16")]; fp16 var_8556_to_fp16 = const()[name = string("op_8556_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_847_cast_fp16, y = var_8556_to_fp16)[name = string("aw_chunk_847_cast_fp16")]; fp16 var_8558_to_fp16 = const()[name = string("op_8558_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_849_cast_fp16, y = var_8558_to_fp16)[name = string("aw_chunk_849_cast_fp16")]; fp16 var_8560_to_fp16 = const()[name = string("op_8560_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_851_cast_fp16, y = var_8560_to_fp16)[name = string("aw_chunk_851_cast_fp16")]; fp16 var_8562_to_fp16 = const()[name = string("op_8562_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_853_cast_fp16, y = var_8562_to_fp16)[name = string("aw_chunk_853_cast_fp16")]; fp16 var_8564_to_fp16 = const()[name = string("op_8564_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_855_cast_fp16, y = var_8564_to_fp16)[name = string("aw_chunk_855_cast_fp16")]; fp16 var_8566_to_fp16 = const()[name = string("op_8566_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_857_cast_fp16, y = var_8566_to_fp16)[name = string("aw_chunk_857_cast_fp16")]; fp16 var_8568_to_fp16 = const()[name = string("op_8568_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_859_cast_fp16, y = var_8568_to_fp16)[name = string("aw_chunk_859_cast_fp16")]; fp16 var_8570_to_fp16 = const()[name = string("op_8570_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_861_cast_fp16, y = var_8570_to_fp16)[name = string("aw_chunk_861_cast_fp16")]; fp16 var_8572_to_fp16 = const()[name = string("op_8572_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_863_cast_fp16, y = var_8572_to_fp16)[name = string("aw_chunk_863_cast_fp16")]; tensor var_8574_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_769_cast_fp16)[name = string("op_8574_cast_fp16")]; tensor var_8575_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_771_cast_fp16)[name = string("op_8575_cast_fp16")]; tensor var_8576_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_773_cast_fp16)[name = string("op_8576_cast_fp16")]; tensor var_8577_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_775_cast_fp16)[name = string("op_8577_cast_fp16")]; tensor var_8578_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_777_cast_fp16)[name = string("op_8578_cast_fp16")]; tensor var_8579_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_779_cast_fp16)[name = string("op_8579_cast_fp16")]; tensor var_8580_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_781_cast_fp16)[name = string("op_8580_cast_fp16")]; tensor var_8581_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_783_cast_fp16)[name = string("op_8581_cast_fp16")]; tensor var_8582_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_785_cast_fp16)[name = string("op_8582_cast_fp16")]; tensor var_8583_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_787_cast_fp16)[name = string("op_8583_cast_fp16")]; tensor var_8584_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_789_cast_fp16)[name = string("op_8584_cast_fp16")]; tensor var_8585_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_791_cast_fp16)[name = string("op_8585_cast_fp16")]; tensor var_8586_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_793_cast_fp16)[name = string("op_8586_cast_fp16")]; tensor var_8587_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_795_cast_fp16)[name = string("op_8587_cast_fp16")]; tensor var_8588_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_797_cast_fp16)[name = string("op_8588_cast_fp16")]; tensor var_8589_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_799_cast_fp16)[name = string("op_8589_cast_fp16")]; tensor var_8590_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_801_cast_fp16)[name = string("op_8590_cast_fp16")]; tensor var_8591_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_803_cast_fp16)[name = string("op_8591_cast_fp16")]; tensor var_8592_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_805_cast_fp16)[name = string("op_8592_cast_fp16")]; tensor var_8593_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_807_cast_fp16)[name = string("op_8593_cast_fp16")]; tensor var_8594_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_809_cast_fp16)[name = string("op_8594_cast_fp16")]; tensor var_8595_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_811_cast_fp16)[name = string("op_8595_cast_fp16")]; tensor var_8596_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_813_cast_fp16)[name = string("op_8596_cast_fp16")]; tensor var_8597_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_815_cast_fp16)[name = string("op_8597_cast_fp16")]; tensor var_8598_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_817_cast_fp16)[name = string("op_8598_cast_fp16")]; tensor var_8599_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_819_cast_fp16)[name = string("op_8599_cast_fp16")]; tensor var_8600_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_821_cast_fp16)[name = string("op_8600_cast_fp16")]; tensor var_8601_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_823_cast_fp16)[name = string("op_8601_cast_fp16")]; tensor var_8602_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_825_cast_fp16)[name = string("op_8602_cast_fp16")]; tensor var_8603_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_827_cast_fp16)[name = string("op_8603_cast_fp16")]; tensor var_8604_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_829_cast_fp16)[name = string("op_8604_cast_fp16")]; tensor var_8605_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_831_cast_fp16)[name = string("op_8605_cast_fp16")]; tensor var_8606_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_833_cast_fp16)[name = string("op_8606_cast_fp16")]; tensor var_8607_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_835_cast_fp16)[name = string("op_8607_cast_fp16")]; tensor var_8608_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_837_cast_fp16)[name = string("op_8608_cast_fp16")]; tensor var_8609_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_839_cast_fp16)[name = string("op_8609_cast_fp16")]; tensor var_8610_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_841_cast_fp16)[name = string("op_8610_cast_fp16")]; tensor var_8611_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_843_cast_fp16)[name = string("op_8611_cast_fp16")]; tensor var_8612_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_845_cast_fp16)[name = string("op_8612_cast_fp16")]; tensor var_8613_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_847_cast_fp16)[name = string("op_8613_cast_fp16")]; tensor var_8614_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_849_cast_fp16)[name = string("op_8614_cast_fp16")]; tensor var_8615_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_851_cast_fp16)[name = string("op_8615_cast_fp16")]; tensor var_8616_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_853_cast_fp16)[name = string("op_8616_cast_fp16")]; tensor var_8617_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_855_cast_fp16)[name = string("op_8617_cast_fp16")]; tensor var_8618_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_857_cast_fp16)[name = string("op_8618_cast_fp16")]; tensor var_8619_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_859_cast_fp16)[name = string("op_8619_cast_fp16")]; tensor var_8620_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_861_cast_fp16)[name = string("op_8620_cast_fp16")]; tensor var_8621_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_863_cast_fp16)[name = string("op_8621_cast_fp16")]; string var_8623_equation_0 = const()[name = string("op_8623_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8623_cast_fp16 = einsum(equation = var_8623_equation_0, values = (var_8335_cast_fp16, var_8574_cast_fp16))[name = string("op_8623_cast_fp16")]; string var_8625_equation_0 = const()[name = string("op_8625_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8625_cast_fp16 = einsum(equation = var_8625_equation_0, values = (var_8335_cast_fp16, var_8575_cast_fp16))[name = string("op_8625_cast_fp16")]; string var_8627_equation_0 = const()[name = string("op_8627_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8627_cast_fp16 = einsum(equation = var_8627_equation_0, values = (var_8335_cast_fp16, var_8576_cast_fp16))[name = string("op_8627_cast_fp16")]; string var_8629_equation_0 = const()[name = string("op_8629_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8629_cast_fp16 = einsum(equation = var_8629_equation_0, values = (var_8335_cast_fp16, var_8577_cast_fp16))[name = string("op_8629_cast_fp16")]; string var_8631_equation_0 = const()[name = string("op_8631_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8631_cast_fp16 = einsum(equation = var_8631_equation_0, values = (var_8339_cast_fp16, var_8578_cast_fp16))[name = string("op_8631_cast_fp16")]; string var_8633_equation_0 = const()[name = string("op_8633_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8633_cast_fp16 = einsum(equation = var_8633_equation_0, values = (var_8339_cast_fp16, var_8579_cast_fp16))[name = string("op_8633_cast_fp16")]; string var_8635_equation_0 = const()[name = string("op_8635_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8635_cast_fp16 = einsum(equation = var_8635_equation_0, values = (var_8339_cast_fp16, var_8580_cast_fp16))[name = string("op_8635_cast_fp16")]; string var_8637_equation_0 = const()[name = string("op_8637_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8637_cast_fp16 = einsum(equation = var_8637_equation_0, values = (var_8339_cast_fp16, var_8581_cast_fp16))[name = string("op_8637_cast_fp16")]; string var_8639_equation_0 = const()[name = string("op_8639_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8639_cast_fp16 = einsum(equation = var_8639_equation_0, values = (var_8343_cast_fp16, var_8582_cast_fp16))[name = string("op_8639_cast_fp16")]; string var_8641_equation_0 = const()[name = string("op_8641_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8641_cast_fp16 = einsum(equation = var_8641_equation_0, values = (var_8343_cast_fp16, var_8583_cast_fp16))[name = string("op_8641_cast_fp16")]; string var_8643_equation_0 = const()[name = string("op_8643_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8643_cast_fp16 = einsum(equation = var_8643_equation_0, values = (var_8343_cast_fp16, var_8584_cast_fp16))[name = string("op_8643_cast_fp16")]; string var_8645_equation_0 = const()[name = string("op_8645_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8645_cast_fp16 = einsum(equation = var_8645_equation_0, values = (var_8343_cast_fp16, var_8585_cast_fp16))[name = string("op_8645_cast_fp16")]; string var_8647_equation_0 = const()[name = string("op_8647_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8647_cast_fp16 = einsum(equation = var_8647_equation_0, values = (var_8347_cast_fp16, var_8586_cast_fp16))[name = string("op_8647_cast_fp16")]; string var_8649_equation_0 = const()[name = string("op_8649_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8649_cast_fp16 = einsum(equation = var_8649_equation_0, values = (var_8347_cast_fp16, var_8587_cast_fp16))[name = string("op_8649_cast_fp16")]; string var_8651_equation_0 = const()[name = string("op_8651_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8651_cast_fp16 = einsum(equation = var_8651_equation_0, values = (var_8347_cast_fp16, var_8588_cast_fp16))[name = string("op_8651_cast_fp16")]; string var_8653_equation_0 = const()[name = string("op_8653_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8653_cast_fp16 = einsum(equation = var_8653_equation_0, values = (var_8347_cast_fp16, var_8589_cast_fp16))[name = string("op_8653_cast_fp16")]; string var_8655_equation_0 = const()[name = string("op_8655_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8655_cast_fp16 = einsum(equation = var_8655_equation_0, values = (var_8351_cast_fp16, var_8590_cast_fp16))[name = string("op_8655_cast_fp16")]; string var_8657_equation_0 = const()[name = string("op_8657_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8657_cast_fp16 = einsum(equation = var_8657_equation_0, values = (var_8351_cast_fp16, var_8591_cast_fp16))[name = string("op_8657_cast_fp16")]; string var_8659_equation_0 = const()[name = string("op_8659_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8659_cast_fp16 = einsum(equation = var_8659_equation_0, values = (var_8351_cast_fp16, var_8592_cast_fp16))[name = string("op_8659_cast_fp16")]; string var_8661_equation_0 = const()[name = string("op_8661_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8661_cast_fp16 = einsum(equation = var_8661_equation_0, values = (var_8351_cast_fp16, var_8593_cast_fp16))[name = string("op_8661_cast_fp16")]; string var_8663_equation_0 = const()[name = string("op_8663_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8663_cast_fp16 = einsum(equation = var_8663_equation_0, values = (var_8355_cast_fp16, var_8594_cast_fp16))[name = string("op_8663_cast_fp16")]; string var_8665_equation_0 = const()[name = string("op_8665_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8665_cast_fp16 = einsum(equation = var_8665_equation_0, values = (var_8355_cast_fp16, var_8595_cast_fp16))[name = string("op_8665_cast_fp16")]; string var_8667_equation_0 = const()[name = string("op_8667_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8667_cast_fp16 = einsum(equation = var_8667_equation_0, values = (var_8355_cast_fp16, var_8596_cast_fp16))[name = string("op_8667_cast_fp16")]; string var_8669_equation_0 = const()[name = string("op_8669_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8669_cast_fp16 = einsum(equation = var_8669_equation_0, values = (var_8355_cast_fp16, var_8597_cast_fp16))[name = string("op_8669_cast_fp16")]; string var_8671_equation_0 = const()[name = string("op_8671_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8671_cast_fp16 = einsum(equation = var_8671_equation_0, values = (var_8359_cast_fp16, var_8598_cast_fp16))[name = string("op_8671_cast_fp16")]; string var_8673_equation_0 = const()[name = string("op_8673_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8673_cast_fp16 = einsum(equation = var_8673_equation_0, values = (var_8359_cast_fp16, var_8599_cast_fp16))[name = string("op_8673_cast_fp16")]; string var_8675_equation_0 = const()[name = string("op_8675_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8675_cast_fp16 = einsum(equation = var_8675_equation_0, values = (var_8359_cast_fp16, var_8600_cast_fp16))[name = string("op_8675_cast_fp16")]; string var_8677_equation_0 = const()[name = string("op_8677_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8677_cast_fp16 = einsum(equation = var_8677_equation_0, values = (var_8359_cast_fp16, var_8601_cast_fp16))[name = string("op_8677_cast_fp16")]; string var_8679_equation_0 = const()[name = string("op_8679_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8679_cast_fp16 = einsum(equation = var_8679_equation_0, values = (var_8363_cast_fp16, var_8602_cast_fp16))[name = string("op_8679_cast_fp16")]; string var_8681_equation_0 = const()[name = string("op_8681_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8681_cast_fp16 = einsum(equation = var_8681_equation_0, values = (var_8363_cast_fp16, var_8603_cast_fp16))[name = string("op_8681_cast_fp16")]; string var_8683_equation_0 = const()[name = string("op_8683_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8683_cast_fp16 = einsum(equation = var_8683_equation_0, values = (var_8363_cast_fp16, var_8604_cast_fp16))[name = string("op_8683_cast_fp16")]; string var_8685_equation_0 = const()[name = string("op_8685_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8685_cast_fp16 = einsum(equation = var_8685_equation_0, values = (var_8363_cast_fp16, var_8605_cast_fp16))[name = string("op_8685_cast_fp16")]; string var_8687_equation_0 = const()[name = string("op_8687_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8687_cast_fp16 = einsum(equation = var_8687_equation_0, values = (var_8367_cast_fp16, var_8606_cast_fp16))[name = string("op_8687_cast_fp16")]; string var_8689_equation_0 = const()[name = string("op_8689_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8689_cast_fp16 = einsum(equation = var_8689_equation_0, values = (var_8367_cast_fp16, var_8607_cast_fp16))[name = string("op_8689_cast_fp16")]; string var_8691_equation_0 = const()[name = string("op_8691_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8691_cast_fp16 = einsum(equation = var_8691_equation_0, values = (var_8367_cast_fp16, var_8608_cast_fp16))[name = string("op_8691_cast_fp16")]; string var_8693_equation_0 = const()[name = string("op_8693_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8693_cast_fp16 = einsum(equation = var_8693_equation_0, values = (var_8367_cast_fp16, var_8609_cast_fp16))[name = string("op_8693_cast_fp16")]; string var_8695_equation_0 = const()[name = string("op_8695_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8695_cast_fp16 = einsum(equation = var_8695_equation_0, values = (var_8371_cast_fp16, var_8610_cast_fp16))[name = string("op_8695_cast_fp16")]; string var_8697_equation_0 = const()[name = string("op_8697_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8697_cast_fp16 = einsum(equation = var_8697_equation_0, values = (var_8371_cast_fp16, var_8611_cast_fp16))[name = string("op_8697_cast_fp16")]; string var_8699_equation_0 = const()[name = string("op_8699_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8699_cast_fp16 = einsum(equation = var_8699_equation_0, values = (var_8371_cast_fp16, var_8612_cast_fp16))[name = string("op_8699_cast_fp16")]; string var_8701_equation_0 = const()[name = string("op_8701_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8701_cast_fp16 = einsum(equation = var_8701_equation_0, values = (var_8371_cast_fp16, var_8613_cast_fp16))[name = string("op_8701_cast_fp16")]; string var_8703_equation_0 = const()[name = string("op_8703_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8703_cast_fp16 = einsum(equation = var_8703_equation_0, values = (var_8375_cast_fp16, var_8614_cast_fp16))[name = string("op_8703_cast_fp16")]; string var_8705_equation_0 = const()[name = string("op_8705_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8705_cast_fp16 = einsum(equation = var_8705_equation_0, values = (var_8375_cast_fp16, var_8615_cast_fp16))[name = string("op_8705_cast_fp16")]; string var_8707_equation_0 = const()[name = string("op_8707_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8707_cast_fp16 = einsum(equation = var_8707_equation_0, values = (var_8375_cast_fp16, var_8616_cast_fp16))[name = string("op_8707_cast_fp16")]; string var_8709_equation_0 = const()[name = string("op_8709_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8709_cast_fp16 = einsum(equation = var_8709_equation_0, values = (var_8375_cast_fp16, var_8617_cast_fp16))[name = string("op_8709_cast_fp16")]; string var_8711_equation_0 = const()[name = string("op_8711_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8711_cast_fp16 = einsum(equation = var_8711_equation_0, values = (var_8379_cast_fp16, var_8618_cast_fp16))[name = string("op_8711_cast_fp16")]; string var_8713_equation_0 = const()[name = string("op_8713_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8713_cast_fp16 = einsum(equation = var_8713_equation_0, values = (var_8379_cast_fp16, var_8619_cast_fp16))[name = string("op_8713_cast_fp16")]; string var_8715_equation_0 = const()[name = string("op_8715_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8715_cast_fp16 = einsum(equation = var_8715_equation_0, values = (var_8379_cast_fp16, var_8620_cast_fp16))[name = string("op_8715_cast_fp16")]; string var_8717_equation_0 = const()[name = string("op_8717_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_8717_cast_fp16 = einsum(equation = var_8717_equation_0, values = (var_8379_cast_fp16, var_8621_cast_fp16))[name = string("op_8717_cast_fp16")]; bool var_8719_interleave_0 = const()[name = string("op_8719_interleave_0"), val = bool(false)]; tensor var_8719_cast_fp16 = concat(axis = var_7830, interleave = var_8719_interleave_0, values = (var_8623_cast_fp16, var_8625_cast_fp16, var_8627_cast_fp16, var_8629_cast_fp16))[name = string("op_8719_cast_fp16")]; bool var_8721_interleave_0 = const()[name = string("op_8721_interleave_0"), val = bool(false)]; tensor var_8721_cast_fp16 = concat(axis = var_7830, interleave = var_8721_interleave_0, values = (var_8631_cast_fp16, var_8633_cast_fp16, var_8635_cast_fp16, var_8637_cast_fp16))[name = string("op_8721_cast_fp16")]; bool var_8723_interleave_0 = const()[name = string("op_8723_interleave_0"), val = bool(false)]; tensor var_8723_cast_fp16 = concat(axis = var_7830, interleave = var_8723_interleave_0, values = (var_8639_cast_fp16, var_8641_cast_fp16, var_8643_cast_fp16, var_8645_cast_fp16))[name = string("op_8723_cast_fp16")]; bool var_8725_interleave_0 = const()[name = string("op_8725_interleave_0"), val = bool(false)]; tensor var_8725_cast_fp16 = concat(axis = var_7830, interleave = var_8725_interleave_0, values = (var_8647_cast_fp16, var_8649_cast_fp16, var_8651_cast_fp16, var_8653_cast_fp16))[name = string("op_8725_cast_fp16")]; bool var_8727_interleave_0 = const()[name = string("op_8727_interleave_0"), val = bool(false)]; tensor var_8727_cast_fp16 = concat(axis = var_7830, interleave = var_8727_interleave_0, values = (var_8655_cast_fp16, var_8657_cast_fp16, var_8659_cast_fp16, var_8661_cast_fp16))[name = string("op_8727_cast_fp16")]; bool var_8729_interleave_0 = const()[name = string("op_8729_interleave_0"), val = bool(false)]; tensor var_8729_cast_fp16 = concat(axis = var_7830, interleave = var_8729_interleave_0, values = (var_8663_cast_fp16, var_8665_cast_fp16, var_8667_cast_fp16, var_8669_cast_fp16))[name = string("op_8729_cast_fp16")]; bool var_8731_interleave_0 = const()[name = string("op_8731_interleave_0"), val = bool(false)]; tensor var_8731_cast_fp16 = concat(axis = var_7830, interleave = var_8731_interleave_0, values = (var_8671_cast_fp16, var_8673_cast_fp16, var_8675_cast_fp16, var_8677_cast_fp16))[name = string("op_8731_cast_fp16")]; bool var_8733_interleave_0 = const()[name = string("op_8733_interleave_0"), val = bool(false)]; tensor var_8733_cast_fp16 = concat(axis = var_7830, interleave = var_8733_interleave_0, values = (var_8679_cast_fp16, var_8681_cast_fp16, var_8683_cast_fp16, var_8685_cast_fp16))[name = string("op_8733_cast_fp16")]; bool var_8735_interleave_0 = const()[name = string("op_8735_interleave_0"), val = bool(false)]; tensor var_8735_cast_fp16 = concat(axis = var_7830, interleave = var_8735_interleave_0, values = (var_8687_cast_fp16, var_8689_cast_fp16, var_8691_cast_fp16, var_8693_cast_fp16))[name = string("op_8735_cast_fp16")]; bool var_8737_interleave_0 = const()[name = string("op_8737_interleave_0"), val = bool(false)]; tensor var_8737_cast_fp16 = concat(axis = var_7830, interleave = var_8737_interleave_0, values = (var_8695_cast_fp16, var_8697_cast_fp16, var_8699_cast_fp16, var_8701_cast_fp16))[name = string("op_8737_cast_fp16")]; bool var_8739_interleave_0 = const()[name = string("op_8739_interleave_0"), val = bool(false)]; tensor var_8739_cast_fp16 = concat(axis = var_7830, interleave = var_8739_interleave_0, values = (var_8703_cast_fp16, var_8705_cast_fp16, var_8707_cast_fp16, var_8709_cast_fp16))[name = string("op_8739_cast_fp16")]; bool var_8741_interleave_0 = const()[name = string("op_8741_interleave_0"), val = bool(false)]; tensor var_8741_cast_fp16 = concat(axis = var_7830, interleave = var_8741_interleave_0, values = (var_8711_cast_fp16, var_8713_cast_fp16, var_8715_cast_fp16, var_8717_cast_fp16))[name = string("op_8741_cast_fp16")]; bool input_65_interleave_0 = const()[name = string("input_65_interleave_0"), val = bool(false)]; tensor input_65_cast_fp16 = concat(axis = var_7847, interleave = input_65_interleave_0, values = (var_8719_cast_fp16, var_8721_cast_fp16, var_8723_cast_fp16, var_8725_cast_fp16, var_8727_cast_fp16, var_8729_cast_fp16, var_8731_cast_fp16, var_8733_cast_fp16, var_8735_cast_fp16, var_8737_cast_fp16, var_8739_cast_fp16, var_8741_cast_fp16))[name = string("input_65_cast_fp16")]; string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123165120)))]; tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124344832)))]; tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; fp16 var_8760_to_fp16 = const()[name = string("op_8760_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_8760_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124346432)))]; tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124348032)))]; fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124349632)))]; tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129068288)))]; tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129074496)))]; tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133793152)))]; tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; int32 var_8789 = const()[name = string("op_8789"), val = int32(3)]; int32 var_8806 = const()[name = string("op_8806"), val = int32(1)]; tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; fp16 var_8823_to_fp16 = const()[name = string("op_8823_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_8823_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133794752)))]; tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133796352)))]; fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133797952)))]; tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134977664)))]; tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134979264)))]; tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136158976)))]; tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137338688)))]; tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; tensor var_8861_begin_0 = const()[name = string("op_8861_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8861_end_0 = const()[name = string("op_8861_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8861_end_mask_0 = const()[name = string("op_8861_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8861_cast_fp16 = slice_by_index(begin = var_8861_begin_0, end = var_8861_end_0, end_mask = var_8861_end_mask_0, x = query_19_cast_fp16)[name = string("op_8861_cast_fp16")]; tensor var_8865_begin_0 = const()[name = string("op_8865_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8865_end_0 = const()[name = string("op_8865_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_8865_end_mask_0 = const()[name = string("op_8865_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8865_cast_fp16 = slice_by_index(begin = var_8865_begin_0, end = var_8865_end_0, end_mask = var_8865_end_mask_0, x = query_19_cast_fp16)[name = string("op_8865_cast_fp16")]; tensor var_8869_begin_0 = const()[name = string("op_8869_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8869_end_0 = const()[name = string("op_8869_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_8869_end_mask_0 = const()[name = string("op_8869_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8869_cast_fp16 = slice_by_index(begin = var_8869_begin_0, end = var_8869_end_0, end_mask = var_8869_end_mask_0, x = query_19_cast_fp16)[name = string("op_8869_cast_fp16")]; tensor var_8873_begin_0 = const()[name = string("op_8873_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8873_end_0 = const()[name = string("op_8873_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_8873_end_mask_0 = const()[name = string("op_8873_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8873_cast_fp16 = slice_by_index(begin = var_8873_begin_0, end = var_8873_end_0, end_mask = var_8873_end_mask_0, x = query_19_cast_fp16)[name = string("op_8873_cast_fp16")]; tensor var_8877_begin_0 = const()[name = string("op_8877_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8877_end_0 = const()[name = string("op_8877_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_8877_end_mask_0 = const()[name = string("op_8877_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8877_cast_fp16 = slice_by_index(begin = var_8877_begin_0, end = var_8877_end_0, end_mask = var_8877_end_mask_0, x = query_19_cast_fp16)[name = string("op_8877_cast_fp16")]; tensor var_8881_begin_0 = const()[name = string("op_8881_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8881_end_0 = const()[name = string("op_8881_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_8881_end_mask_0 = const()[name = string("op_8881_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8881_cast_fp16 = slice_by_index(begin = var_8881_begin_0, end = var_8881_end_0, end_mask = var_8881_end_mask_0, x = query_19_cast_fp16)[name = string("op_8881_cast_fp16")]; tensor var_8885_begin_0 = const()[name = string("op_8885_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8885_end_0 = const()[name = string("op_8885_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_8885_end_mask_0 = const()[name = string("op_8885_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8885_cast_fp16 = slice_by_index(begin = var_8885_begin_0, end = var_8885_end_0, end_mask = var_8885_end_mask_0, x = query_19_cast_fp16)[name = string("op_8885_cast_fp16")]; tensor var_8889_begin_0 = const()[name = string("op_8889_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8889_end_0 = const()[name = string("op_8889_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_8889_end_mask_0 = const()[name = string("op_8889_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8889_cast_fp16 = slice_by_index(begin = var_8889_begin_0, end = var_8889_end_0, end_mask = var_8889_end_mask_0, x = query_19_cast_fp16)[name = string("op_8889_cast_fp16")]; tensor var_8893_begin_0 = const()[name = string("op_8893_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8893_end_0 = const()[name = string("op_8893_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_8893_end_mask_0 = const()[name = string("op_8893_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8893_cast_fp16 = slice_by_index(begin = var_8893_begin_0, end = var_8893_end_0, end_mask = var_8893_end_mask_0, x = query_19_cast_fp16)[name = string("op_8893_cast_fp16")]; tensor var_8897_begin_0 = const()[name = string("op_8897_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8897_end_0 = const()[name = string("op_8897_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_8897_end_mask_0 = const()[name = string("op_8897_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8897_cast_fp16 = slice_by_index(begin = var_8897_begin_0, end = var_8897_end_0, end_mask = var_8897_end_mask_0, x = query_19_cast_fp16)[name = string("op_8897_cast_fp16")]; tensor var_8901_begin_0 = const()[name = string("op_8901_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8901_end_0 = const()[name = string("op_8901_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_8901_end_mask_0 = const()[name = string("op_8901_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8901_cast_fp16 = slice_by_index(begin = var_8901_begin_0, end = var_8901_end_0, end_mask = var_8901_end_mask_0, x = query_19_cast_fp16)[name = string("op_8901_cast_fp16")]; tensor var_8905_begin_0 = const()[name = string("op_8905_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8905_end_0 = const()[name = string("op_8905_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_8905_end_mask_0 = const()[name = string("op_8905_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8905_cast_fp16 = slice_by_index(begin = var_8905_begin_0, end = var_8905_end_0, end_mask = var_8905_end_mask_0, x = query_19_cast_fp16)[name = string("op_8905_cast_fp16")]; tensor var_8914_begin_0 = const()[name = string("op_8914_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8914_end_0 = const()[name = string("op_8914_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8914_end_mask_0 = const()[name = string("op_8914_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8914_cast_fp16 = slice_by_index(begin = var_8914_begin_0, end = var_8914_end_0, end_mask = var_8914_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8914_cast_fp16")]; tensor var_8921_begin_0 = const()[name = string("op_8921_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8921_end_0 = const()[name = string("op_8921_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8921_end_mask_0 = const()[name = string("op_8921_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8921_cast_fp16 = slice_by_index(begin = var_8921_begin_0, end = var_8921_end_0, end_mask = var_8921_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8921_cast_fp16")]; tensor var_8928_begin_0 = const()[name = string("op_8928_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8928_end_0 = const()[name = string("op_8928_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8928_end_mask_0 = const()[name = string("op_8928_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8928_cast_fp16 = slice_by_index(begin = var_8928_begin_0, end = var_8928_end_0, end_mask = var_8928_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8928_cast_fp16")]; tensor var_8935_begin_0 = const()[name = string("op_8935_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8935_end_0 = const()[name = string("op_8935_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8935_end_mask_0 = const()[name = string("op_8935_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8935_cast_fp16 = slice_by_index(begin = var_8935_begin_0, end = var_8935_end_0, end_mask = var_8935_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8935_cast_fp16")]; tensor var_8942_begin_0 = const()[name = string("op_8942_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8942_end_0 = const()[name = string("op_8942_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8942_end_mask_0 = const()[name = string("op_8942_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8942_cast_fp16 = slice_by_index(begin = var_8942_begin_0, end = var_8942_end_0, end_mask = var_8942_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8942_cast_fp16")]; tensor var_8949_begin_0 = const()[name = string("op_8949_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8949_end_0 = const()[name = string("op_8949_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8949_end_mask_0 = const()[name = string("op_8949_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8949_cast_fp16 = slice_by_index(begin = var_8949_begin_0, end = var_8949_end_0, end_mask = var_8949_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8949_cast_fp16")]; tensor var_8956_begin_0 = const()[name = string("op_8956_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8956_end_0 = const()[name = string("op_8956_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8956_end_mask_0 = const()[name = string("op_8956_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8956_cast_fp16 = slice_by_index(begin = var_8956_begin_0, end = var_8956_end_0, end_mask = var_8956_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8956_cast_fp16")]; tensor var_8963_begin_0 = const()[name = string("op_8963_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8963_end_0 = const()[name = string("op_8963_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8963_end_mask_0 = const()[name = string("op_8963_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8963_cast_fp16 = slice_by_index(begin = var_8963_begin_0, end = var_8963_end_0, end_mask = var_8963_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8963_cast_fp16")]; tensor var_8970_begin_0 = const()[name = string("op_8970_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8970_end_0 = const()[name = string("op_8970_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8970_end_mask_0 = const()[name = string("op_8970_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8970_cast_fp16 = slice_by_index(begin = var_8970_begin_0, end = var_8970_end_0, end_mask = var_8970_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8970_cast_fp16")]; tensor var_8977_begin_0 = const()[name = string("op_8977_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8977_end_0 = const()[name = string("op_8977_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8977_end_mask_0 = const()[name = string("op_8977_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8977_cast_fp16 = slice_by_index(begin = var_8977_begin_0, end = var_8977_end_0, end_mask = var_8977_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8977_cast_fp16")]; tensor var_8984_begin_0 = const()[name = string("op_8984_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8984_end_0 = const()[name = string("op_8984_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8984_end_mask_0 = const()[name = string("op_8984_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8984_cast_fp16 = slice_by_index(begin = var_8984_begin_0, end = var_8984_end_0, end_mask = var_8984_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8984_cast_fp16")]; tensor var_8991_begin_0 = const()[name = string("op_8991_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8991_end_0 = const()[name = string("op_8991_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8991_end_mask_0 = const()[name = string("op_8991_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8991_cast_fp16 = slice_by_index(begin = var_8991_begin_0, end = var_8991_end_0, end_mask = var_8991_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8991_cast_fp16")]; tensor var_8998_begin_0 = const()[name = string("op_8998_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8998_end_0 = const()[name = string("op_8998_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8998_end_mask_0 = const()[name = string("op_8998_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8998_cast_fp16 = slice_by_index(begin = var_8998_begin_0, end = var_8998_end_0, end_mask = var_8998_end_mask_0, x = var_8873_cast_fp16)[name = string("op_8998_cast_fp16")]; tensor var_9005_begin_0 = const()[name = string("op_9005_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9005_end_0 = const()[name = string("op_9005_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9005_end_mask_0 = const()[name = string("op_9005_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9005_cast_fp16 = slice_by_index(begin = var_9005_begin_0, end = var_9005_end_0, end_mask = var_9005_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9005_cast_fp16")]; tensor var_9012_begin_0 = const()[name = string("op_9012_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9012_end_0 = const()[name = string("op_9012_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9012_end_mask_0 = const()[name = string("op_9012_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9012_cast_fp16 = slice_by_index(begin = var_9012_begin_0, end = var_9012_end_0, end_mask = var_9012_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9012_cast_fp16")]; tensor var_9019_begin_0 = const()[name = string("op_9019_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9019_end_0 = const()[name = string("op_9019_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9019_end_mask_0 = const()[name = string("op_9019_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9019_cast_fp16 = slice_by_index(begin = var_9019_begin_0, end = var_9019_end_0, end_mask = var_9019_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9019_cast_fp16")]; tensor var_9026_begin_0 = const()[name = string("op_9026_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9026_end_0 = const()[name = string("op_9026_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9026_end_mask_0 = const()[name = string("op_9026_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9026_cast_fp16 = slice_by_index(begin = var_9026_begin_0, end = var_9026_end_0, end_mask = var_9026_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9026_cast_fp16")]; tensor var_9033_begin_0 = const()[name = string("op_9033_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9033_end_0 = const()[name = string("op_9033_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9033_end_mask_0 = const()[name = string("op_9033_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9033_cast_fp16 = slice_by_index(begin = var_9033_begin_0, end = var_9033_end_0, end_mask = var_9033_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9033_cast_fp16")]; tensor var_9040_begin_0 = const()[name = string("op_9040_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9040_end_0 = const()[name = string("op_9040_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9040_end_mask_0 = const()[name = string("op_9040_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9040_cast_fp16 = slice_by_index(begin = var_9040_begin_0, end = var_9040_end_0, end_mask = var_9040_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9040_cast_fp16")]; tensor var_9047_begin_0 = const()[name = string("op_9047_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9047_end_0 = const()[name = string("op_9047_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9047_end_mask_0 = const()[name = string("op_9047_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9047_cast_fp16 = slice_by_index(begin = var_9047_begin_0, end = var_9047_end_0, end_mask = var_9047_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9047_cast_fp16")]; tensor var_9054_begin_0 = const()[name = string("op_9054_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9054_end_0 = const()[name = string("op_9054_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9054_end_mask_0 = const()[name = string("op_9054_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9054_cast_fp16 = slice_by_index(begin = var_9054_begin_0, end = var_9054_end_0, end_mask = var_9054_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9054_cast_fp16")]; tensor var_9061_begin_0 = const()[name = string("op_9061_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9061_end_0 = const()[name = string("op_9061_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9061_end_mask_0 = const()[name = string("op_9061_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9061_cast_fp16 = slice_by_index(begin = var_9061_begin_0, end = var_9061_end_0, end_mask = var_9061_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9061_cast_fp16")]; tensor var_9068_begin_0 = const()[name = string("op_9068_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9068_end_0 = const()[name = string("op_9068_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9068_end_mask_0 = const()[name = string("op_9068_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9068_cast_fp16 = slice_by_index(begin = var_9068_begin_0, end = var_9068_end_0, end_mask = var_9068_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9068_cast_fp16")]; tensor var_9075_begin_0 = const()[name = string("op_9075_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9075_end_0 = const()[name = string("op_9075_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9075_end_mask_0 = const()[name = string("op_9075_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9075_cast_fp16 = slice_by_index(begin = var_9075_begin_0, end = var_9075_end_0, end_mask = var_9075_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9075_cast_fp16")]; tensor var_9082_begin_0 = const()[name = string("op_9082_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9082_end_0 = const()[name = string("op_9082_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9082_end_mask_0 = const()[name = string("op_9082_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9082_cast_fp16 = slice_by_index(begin = var_9082_begin_0, end = var_9082_end_0, end_mask = var_9082_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9082_cast_fp16")]; tensor var_9089_begin_0 = const()[name = string("op_9089_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9089_end_0 = const()[name = string("op_9089_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9089_end_mask_0 = const()[name = string("op_9089_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9089_cast_fp16 = slice_by_index(begin = var_9089_begin_0, end = var_9089_end_0, end_mask = var_9089_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9089_cast_fp16")]; tensor var_9096_begin_0 = const()[name = string("op_9096_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9096_end_0 = const()[name = string("op_9096_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9096_end_mask_0 = const()[name = string("op_9096_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9096_cast_fp16 = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9096_cast_fp16")]; tensor var_9103_begin_0 = const()[name = string("op_9103_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9103_end_0 = const()[name = string("op_9103_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9103_end_mask_0 = const()[name = string("op_9103_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9103_cast_fp16 = slice_by_index(begin = var_9103_begin_0, end = var_9103_end_0, end_mask = var_9103_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9103_cast_fp16")]; tensor var_9110_begin_0 = const()[name = string("op_9110_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9110_end_0 = const()[name = string("op_9110_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9110_end_mask_0 = const()[name = string("op_9110_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9110_cast_fp16 = slice_by_index(begin = var_9110_begin_0, end = var_9110_end_0, end_mask = var_9110_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9110_cast_fp16")]; tensor var_9117_begin_0 = const()[name = string("op_9117_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9117_end_0 = const()[name = string("op_9117_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9117_end_mask_0 = const()[name = string("op_9117_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9117_cast_fp16 = slice_by_index(begin = var_9117_begin_0, end = var_9117_end_0, end_mask = var_9117_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9117_cast_fp16")]; tensor var_9124_begin_0 = const()[name = string("op_9124_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9124_end_0 = const()[name = string("op_9124_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9124_end_mask_0 = const()[name = string("op_9124_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9124_cast_fp16 = slice_by_index(begin = var_9124_begin_0, end = var_9124_end_0, end_mask = var_9124_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9124_cast_fp16")]; tensor var_9131_begin_0 = const()[name = string("op_9131_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9131_end_0 = const()[name = string("op_9131_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9131_end_mask_0 = const()[name = string("op_9131_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9131_cast_fp16 = slice_by_index(begin = var_9131_begin_0, end = var_9131_end_0, end_mask = var_9131_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9131_cast_fp16")]; tensor var_9138_begin_0 = const()[name = string("op_9138_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9138_end_0 = const()[name = string("op_9138_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9138_end_mask_0 = const()[name = string("op_9138_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9138_cast_fp16 = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9138_cast_fp16")]; tensor var_9145_begin_0 = const()[name = string("op_9145_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9145_end_0 = const()[name = string("op_9145_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9145_end_mask_0 = const()[name = string("op_9145_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9145_cast_fp16 = slice_by_index(begin = var_9145_begin_0, end = var_9145_end_0, end_mask = var_9145_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9145_cast_fp16")]; tensor var_9152_begin_0 = const()[name = string("op_9152_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9152_end_0 = const()[name = string("op_9152_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9152_end_mask_0 = const()[name = string("op_9152_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9152_cast_fp16 = slice_by_index(begin = var_9152_begin_0, end = var_9152_end_0, end_mask = var_9152_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9152_cast_fp16")]; tensor var_9159_begin_0 = const()[name = string("op_9159_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9159_end_0 = const()[name = string("op_9159_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9159_end_mask_0 = const()[name = string("op_9159_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9159_cast_fp16 = slice_by_index(begin = var_9159_begin_0, end = var_9159_end_0, end_mask = var_9159_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9159_cast_fp16")]; tensor var_9166_begin_0 = const()[name = string("op_9166_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9166_end_0 = const()[name = string("op_9166_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9166_end_mask_0 = const()[name = string("op_9166_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9166_cast_fp16 = slice_by_index(begin = var_9166_begin_0, end = var_9166_end_0, end_mask = var_9166_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9166_cast_fp16")]; tensor var_9173_begin_0 = const()[name = string("op_9173_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9173_end_0 = const()[name = string("op_9173_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9173_end_mask_0 = const()[name = string("op_9173_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9173_cast_fp16 = slice_by_index(begin = var_9173_begin_0, end = var_9173_end_0, end_mask = var_9173_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9173_cast_fp16")]; tensor var_9180_begin_0 = const()[name = string("op_9180_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9180_end_0 = const()[name = string("op_9180_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9180_end_mask_0 = const()[name = string("op_9180_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9180_cast_fp16 = slice_by_index(begin = var_9180_begin_0, end = var_9180_end_0, end_mask = var_9180_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9180_cast_fp16")]; tensor var_9187_begin_0 = const()[name = string("op_9187_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9187_end_0 = const()[name = string("op_9187_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9187_end_mask_0 = const()[name = string("op_9187_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9187_cast_fp16 = slice_by_index(begin = var_9187_begin_0, end = var_9187_end_0, end_mask = var_9187_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9187_cast_fp16")]; tensor var_9194_begin_0 = const()[name = string("op_9194_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9194_end_0 = const()[name = string("op_9194_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9194_end_mask_0 = const()[name = string("op_9194_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9194_cast_fp16 = slice_by_index(begin = var_9194_begin_0, end = var_9194_end_0, end_mask = var_9194_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9194_cast_fp16")]; tensor var_9201_begin_0 = const()[name = string("op_9201_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9201_end_0 = const()[name = string("op_9201_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9201_end_mask_0 = const()[name = string("op_9201_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9201_cast_fp16 = slice_by_index(begin = var_9201_begin_0, end = var_9201_end_0, end_mask = var_9201_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9201_cast_fp16")]; tensor var_9208_begin_0 = const()[name = string("op_9208_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9208_end_0 = const()[name = string("op_9208_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9208_end_mask_0 = const()[name = string("op_9208_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9208_cast_fp16 = slice_by_index(begin = var_9208_begin_0, end = var_9208_end_0, end_mask = var_9208_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9208_cast_fp16")]; tensor var_9215_begin_0 = const()[name = string("op_9215_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9215_end_0 = const()[name = string("op_9215_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9215_end_mask_0 = const()[name = string("op_9215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9215_cast_fp16 = slice_by_index(begin = var_9215_begin_0, end = var_9215_end_0, end_mask = var_9215_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9215_cast_fp16")]; tensor var_9222_begin_0 = const()[name = string("op_9222_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9222_end_0 = const()[name = string("op_9222_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9222_end_mask_0 = const()[name = string("op_9222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9222_cast_fp16 = slice_by_index(begin = var_9222_begin_0, end = var_9222_end_0, end_mask = var_9222_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9222_cast_fp16")]; tensor var_9229_begin_0 = const()[name = string("op_9229_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9229_end_0 = const()[name = string("op_9229_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9229_end_mask_0 = const()[name = string("op_9229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9229_cast_fp16 = slice_by_index(begin = var_9229_begin_0, end = var_9229_end_0, end_mask = var_9229_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9229_cast_fp16")]; tensor var_9236_begin_0 = const()[name = string("op_9236_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9236_end_0 = const()[name = string("op_9236_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9236_end_mask_0 = const()[name = string("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9236_cast_fp16")]; tensor var_9243_begin_0 = const()[name = string("op_9243_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9243_end_0 = const()[name = string("op_9243_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9243_end_mask_0 = const()[name = string("op_9243_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9243_cast_fp16 = slice_by_index(begin = var_9243_begin_0, end = var_9243_end_0, end_mask = var_9243_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9243_cast_fp16")]; tensor k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9248_begin_0 = const()[name = string("op_9248_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9248_end_0 = const()[name = string("op_9248_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_9248_end_mask_0 = const()[name = string("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = string("transpose_2")]; tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = k_19_cast_fp16)[name = string("op_9248_cast_fp16")]; tensor var_9252_begin_0 = const()[name = string("op_9252_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_9252_end_0 = const()[name = string("op_9252_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_9252_end_mask_0 = const()[name = string("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = k_19_cast_fp16)[name = string("op_9252_cast_fp16")]; tensor var_9256_begin_0 = const()[name = string("op_9256_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_9256_end_0 = const()[name = string("op_9256_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_9256_end_mask_0 = const()[name = string("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = k_19_cast_fp16)[name = string("op_9256_cast_fp16")]; tensor var_9260_begin_0 = const()[name = string("op_9260_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_9260_end_0 = const()[name = string("op_9260_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_9260_end_mask_0 = const()[name = string("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = k_19_cast_fp16)[name = string("op_9260_cast_fp16")]; tensor var_9264_begin_0 = const()[name = string("op_9264_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_9264_end_0 = const()[name = string("op_9264_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_9264_end_mask_0 = const()[name = string("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = k_19_cast_fp16)[name = string("op_9264_cast_fp16")]; tensor var_9268_begin_0 = const()[name = string("op_9268_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9268_end_0 = const()[name = string("op_9268_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_9268_end_mask_0 = const()[name = string("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = k_19_cast_fp16)[name = string("op_9268_cast_fp16")]; tensor var_9272_begin_0 = const()[name = string("op_9272_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_9272_end_0 = const()[name = string("op_9272_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_9272_end_mask_0 = const()[name = string("op_9272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = k_19_cast_fp16)[name = string("op_9272_cast_fp16")]; tensor var_9276_begin_0 = const()[name = string("op_9276_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_9276_end_0 = const()[name = string("op_9276_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_9276_end_mask_0 = const()[name = string("op_9276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = k_19_cast_fp16)[name = string("op_9276_cast_fp16")]; tensor var_9280_begin_0 = const()[name = string("op_9280_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9280_end_0 = const()[name = string("op_9280_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_9280_end_mask_0 = const()[name = string("op_9280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = k_19_cast_fp16)[name = string("op_9280_cast_fp16")]; tensor var_9284_begin_0 = const()[name = string("op_9284_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_9284_end_0 = const()[name = string("op_9284_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_9284_end_mask_0 = const()[name = string("op_9284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9284_cast_fp16 = slice_by_index(begin = var_9284_begin_0, end = var_9284_end_0, end_mask = var_9284_end_mask_0, x = k_19_cast_fp16)[name = string("op_9284_cast_fp16")]; tensor var_9288_begin_0 = const()[name = string("op_9288_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_9288_end_0 = const()[name = string("op_9288_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_9288_end_mask_0 = const()[name = string("op_9288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9288_cast_fp16 = slice_by_index(begin = var_9288_begin_0, end = var_9288_end_0, end_mask = var_9288_end_mask_0, x = k_19_cast_fp16)[name = string("op_9288_cast_fp16")]; tensor var_9292_begin_0 = const()[name = string("op_9292_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_9292_end_0 = const()[name = string("op_9292_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_9292_end_mask_0 = const()[name = string("op_9292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9292_cast_fp16 = slice_by_index(begin = var_9292_begin_0, end = var_9292_end_0, end_mask = var_9292_end_mask_0, x = k_19_cast_fp16)[name = string("op_9292_cast_fp16")]; tensor var_9294_begin_0 = const()[name = string("op_9294_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9294_end_0 = const()[name = string("op_9294_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9294_end_mask_0 = const()[name = string("op_9294_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9294_cast_fp16 = slice_by_index(begin = var_9294_begin_0, end = var_9294_end_0, end_mask = var_9294_end_mask_0, x = value_19_cast_fp16)[name = string("op_9294_cast_fp16")]; tensor var_9298_begin_0 = const()[name = string("op_9298_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9298_end_0 = const()[name = string("op_9298_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_9298_end_mask_0 = const()[name = string("op_9298_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9298_cast_fp16 = slice_by_index(begin = var_9298_begin_0, end = var_9298_end_0, end_mask = var_9298_end_mask_0, x = value_19_cast_fp16)[name = string("op_9298_cast_fp16")]; tensor var_9302_begin_0 = const()[name = string("op_9302_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9302_end_0 = const()[name = string("op_9302_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_9302_end_mask_0 = const()[name = string("op_9302_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9302_cast_fp16 = slice_by_index(begin = var_9302_begin_0, end = var_9302_end_0, end_mask = var_9302_end_mask_0, x = value_19_cast_fp16)[name = string("op_9302_cast_fp16")]; tensor var_9306_begin_0 = const()[name = string("op_9306_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9306_end_0 = const()[name = string("op_9306_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_9306_end_mask_0 = const()[name = string("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = value_19_cast_fp16)[name = string("op_9306_cast_fp16")]; tensor var_9310_begin_0 = const()[name = string("op_9310_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9310_end_0 = const()[name = string("op_9310_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_9310_end_mask_0 = const()[name = string("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = value_19_cast_fp16)[name = string("op_9310_cast_fp16")]; tensor var_9314_begin_0 = const()[name = string("op_9314_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9314_end_0 = const()[name = string("op_9314_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_9314_end_mask_0 = const()[name = string("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = value_19_cast_fp16)[name = string("op_9314_cast_fp16")]; tensor var_9318_begin_0 = const()[name = string("op_9318_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9318_end_0 = const()[name = string("op_9318_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_9318_end_mask_0 = const()[name = string("op_9318_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9318_cast_fp16 = slice_by_index(begin = var_9318_begin_0, end = var_9318_end_0, end_mask = var_9318_end_mask_0, x = value_19_cast_fp16)[name = string("op_9318_cast_fp16")]; tensor var_9322_begin_0 = const()[name = string("op_9322_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9322_end_0 = const()[name = string("op_9322_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_9322_end_mask_0 = const()[name = string("op_9322_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9322_cast_fp16 = slice_by_index(begin = var_9322_begin_0, end = var_9322_end_0, end_mask = var_9322_end_mask_0, x = value_19_cast_fp16)[name = string("op_9322_cast_fp16")]; tensor var_9326_begin_0 = const()[name = string("op_9326_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9326_end_0 = const()[name = string("op_9326_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_9326_end_mask_0 = const()[name = string("op_9326_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9326_cast_fp16 = slice_by_index(begin = var_9326_begin_0, end = var_9326_end_0, end_mask = var_9326_end_mask_0, x = value_19_cast_fp16)[name = string("op_9326_cast_fp16")]; tensor var_9330_begin_0 = const()[name = string("op_9330_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9330_end_0 = const()[name = string("op_9330_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_9330_end_mask_0 = const()[name = string("op_9330_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9330_cast_fp16 = slice_by_index(begin = var_9330_begin_0, end = var_9330_end_0, end_mask = var_9330_end_mask_0, x = value_19_cast_fp16)[name = string("op_9330_cast_fp16")]; tensor var_9334_begin_0 = const()[name = string("op_9334_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_9334_end_0 = const()[name = string("op_9334_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_9334_end_mask_0 = const()[name = string("op_9334_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9334_cast_fp16 = slice_by_index(begin = var_9334_begin_0, end = var_9334_end_0, end_mask = var_9334_end_mask_0, x = value_19_cast_fp16)[name = string("op_9334_cast_fp16")]; tensor var_9338_begin_0 = const()[name = string("op_9338_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_9338_end_0 = const()[name = string("op_9338_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_9338_end_mask_0 = const()[name = string("op_9338_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9338_cast_fp16 = slice_by_index(begin = var_9338_begin_0, end = var_9338_end_0, end_mask = var_9338_end_mask_0, x = value_19_cast_fp16)[name = string("op_9338_cast_fp16")]; string _SplitHeadsQ__mh_w_865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_865_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_865_equation_0, values = (var_9248_cast_fp16, var_8914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_865_cast_fp16")]; string _SplitHeadsQ__mh_w_867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_867_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_867_equation_0, values = (var_9248_cast_fp16, var_8921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_867_cast_fp16")]; string _SplitHeadsQ__mh_w_869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_869_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_869_equation_0, values = (var_9248_cast_fp16, var_8928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_869_cast_fp16")]; string _SplitHeadsQ__mh_w_871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_871_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_871_equation_0, values = (var_9248_cast_fp16, var_8935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_871_cast_fp16")]; string _SplitHeadsQ__mh_w_873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_873_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_873_equation_0, values = (var_9252_cast_fp16, var_8942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_873_cast_fp16")]; string _SplitHeadsQ__mh_w_875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_875_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_875_equation_0, values = (var_9252_cast_fp16, var_8949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_875_cast_fp16")]; string _SplitHeadsQ__mh_w_877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_877_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_877_equation_0, values = (var_9252_cast_fp16, var_8956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_877_cast_fp16")]; string _SplitHeadsQ__mh_w_879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_879_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_879_equation_0, values = (var_9252_cast_fp16, var_8963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_879_cast_fp16")]; string _SplitHeadsQ__mh_w_881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_881_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_881_equation_0, values = (var_9256_cast_fp16, var_8970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_881_cast_fp16")]; string _SplitHeadsQ__mh_w_883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_883_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_883_equation_0, values = (var_9256_cast_fp16, var_8977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_883_cast_fp16")]; string _SplitHeadsQ__mh_w_885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_885_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_885_equation_0, values = (var_9256_cast_fp16, var_8984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_885_cast_fp16")]; string _SplitHeadsQ__mh_w_887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_887_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_887_equation_0, values = (var_9256_cast_fp16, var_8991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_887_cast_fp16")]; string _SplitHeadsQ__mh_w_889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_889_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_889_equation_0, values = (var_9260_cast_fp16, var_8998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_889_cast_fp16")]; string _SplitHeadsQ__mh_w_891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_891_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_891_equation_0, values = (var_9260_cast_fp16, var_9005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_891_cast_fp16")]; string _SplitHeadsQ__mh_w_893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_893_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_893_equation_0, values = (var_9260_cast_fp16, var_9012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_893_cast_fp16")]; string _SplitHeadsQ__mh_w_895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_895_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_895_equation_0, values = (var_9260_cast_fp16, var_9019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_895_cast_fp16")]; string _SplitHeadsQ__mh_w_897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_897_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_897_equation_0, values = (var_9264_cast_fp16, var_9026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_897_cast_fp16")]; string _SplitHeadsQ__mh_w_899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_899_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_899_equation_0, values = (var_9264_cast_fp16, var_9033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_899_cast_fp16")]; string _SplitHeadsQ__mh_w_901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_901_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_901_equation_0, values = (var_9264_cast_fp16, var_9040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_901_cast_fp16")]; string _SplitHeadsQ__mh_w_903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_903_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_903_equation_0, values = (var_9264_cast_fp16, var_9047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_903_cast_fp16")]; string _SplitHeadsQ__mh_w_905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_905_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_905_equation_0, values = (var_9268_cast_fp16, var_9054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_905_cast_fp16")]; string _SplitHeadsQ__mh_w_907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_907_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_907_equation_0, values = (var_9268_cast_fp16, var_9061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_907_cast_fp16")]; string _SplitHeadsQ__mh_w_909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_909_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_909_equation_0, values = (var_9268_cast_fp16, var_9068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_909_cast_fp16")]; string _SplitHeadsQ__mh_w_911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_911_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_911_equation_0, values = (var_9268_cast_fp16, var_9075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_911_cast_fp16")]; string _SplitHeadsQ__mh_w_913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_913_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_913_equation_0, values = (var_9272_cast_fp16, var_9082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_913_cast_fp16")]; string _SplitHeadsQ__mh_w_915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_915_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_915_equation_0, values = (var_9272_cast_fp16, var_9089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_915_cast_fp16")]; string _SplitHeadsQ__mh_w_917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_917_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_917_equation_0, values = (var_9272_cast_fp16, var_9096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_917_cast_fp16")]; string _SplitHeadsQ__mh_w_919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_919_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_919_equation_0, values = (var_9272_cast_fp16, var_9103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_919_cast_fp16")]; string _SplitHeadsQ__mh_w_921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_921_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_921_equation_0, values = (var_9276_cast_fp16, var_9110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_921_cast_fp16")]; string _SplitHeadsQ__mh_w_923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_923_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_923_equation_0, values = (var_9276_cast_fp16, var_9117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_923_cast_fp16")]; string _SplitHeadsQ__mh_w_925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_925_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_925_equation_0, values = (var_9276_cast_fp16, var_9124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_925_cast_fp16")]; string _SplitHeadsQ__mh_w_927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_927_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_927_equation_0, values = (var_9276_cast_fp16, var_9131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_927_cast_fp16")]; string _SplitHeadsQ__mh_w_929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_929_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_929_equation_0, values = (var_9280_cast_fp16, var_9138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_929_cast_fp16")]; string _SplitHeadsQ__mh_w_931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_931_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_931_equation_0, values = (var_9280_cast_fp16, var_9145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_931_cast_fp16")]; string _SplitHeadsQ__mh_w_933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_933_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_933_equation_0, values = (var_9280_cast_fp16, var_9152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_933_cast_fp16")]; string _SplitHeadsQ__mh_w_935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_935_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_935_equation_0, values = (var_9280_cast_fp16, var_9159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_935_cast_fp16")]; string _SplitHeadsQ__mh_w_937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_937_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_937_equation_0, values = (var_9284_cast_fp16, var_9166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_937_cast_fp16")]; string _SplitHeadsQ__mh_w_939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_939_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_939_equation_0, values = (var_9284_cast_fp16, var_9173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_939_cast_fp16")]; string _SplitHeadsQ__mh_w_941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_941_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_941_equation_0, values = (var_9284_cast_fp16, var_9180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_941_cast_fp16")]; string _SplitHeadsQ__mh_w_943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_943_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_943_equation_0, values = (var_9284_cast_fp16, var_9187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_943_cast_fp16")]; string _SplitHeadsQ__mh_w_945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_945_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_945_equation_0, values = (var_9288_cast_fp16, var_9194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_945_cast_fp16")]; string _SplitHeadsQ__mh_w_947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_947_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_947_equation_0, values = (var_9288_cast_fp16, var_9201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_947_cast_fp16")]; string _SplitHeadsQ__mh_w_949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_949_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_949_equation_0, values = (var_9288_cast_fp16, var_9208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_949_cast_fp16")]; string _SplitHeadsQ__mh_w_951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_951_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_951_equation_0, values = (var_9288_cast_fp16, var_9215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_951_cast_fp16")]; string _SplitHeadsQ__mh_w_953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_953_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_953_equation_0, values = (var_9292_cast_fp16, var_9222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_953_cast_fp16")]; string _SplitHeadsQ__mh_w_955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_955_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_955_equation_0, values = (var_9292_cast_fp16, var_9229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_955_cast_fp16")]; string _SplitHeadsQ__mh_w_957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_957_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_957_equation_0, values = (var_9292_cast_fp16, var_9236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_957_cast_fp16")]; string _SplitHeadsQ__mh_w_959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_959_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_959_equation_0, values = (var_9292_cast_fp16, var_9243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_959_cast_fp16")]; fp16 var_9437_to_fp16 = const()[name = string("op_9437_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_865_cast_fp16, y = var_9437_to_fp16)[name = string("aw_chunk_865_cast_fp16")]; fp16 var_9439_to_fp16 = const()[name = string("op_9439_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_867_cast_fp16, y = var_9439_to_fp16)[name = string("aw_chunk_867_cast_fp16")]; fp16 var_9441_to_fp16 = const()[name = string("op_9441_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_869_cast_fp16, y = var_9441_to_fp16)[name = string("aw_chunk_869_cast_fp16")]; fp16 var_9443_to_fp16 = const()[name = string("op_9443_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_871_cast_fp16, y = var_9443_to_fp16)[name = string("aw_chunk_871_cast_fp16")]; fp16 var_9445_to_fp16 = const()[name = string("op_9445_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_873_cast_fp16, y = var_9445_to_fp16)[name = string("aw_chunk_873_cast_fp16")]; fp16 var_9447_to_fp16 = const()[name = string("op_9447_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_875_cast_fp16, y = var_9447_to_fp16)[name = string("aw_chunk_875_cast_fp16")]; fp16 var_9449_to_fp16 = const()[name = string("op_9449_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_877_cast_fp16, y = var_9449_to_fp16)[name = string("aw_chunk_877_cast_fp16")]; fp16 var_9451_to_fp16 = const()[name = string("op_9451_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_879_cast_fp16, y = var_9451_to_fp16)[name = string("aw_chunk_879_cast_fp16")]; fp16 var_9453_to_fp16 = const()[name = string("op_9453_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_881_cast_fp16, y = var_9453_to_fp16)[name = string("aw_chunk_881_cast_fp16")]; fp16 var_9455_to_fp16 = const()[name = string("op_9455_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_883_cast_fp16, y = var_9455_to_fp16)[name = string("aw_chunk_883_cast_fp16")]; fp16 var_9457_to_fp16 = const()[name = string("op_9457_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_885_cast_fp16, y = var_9457_to_fp16)[name = string("aw_chunk_885_cast_fp16")]; fp16 var_9459_to_fp16 = const()[name = string("op_9459_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_887_cast_fp16, y = var_9459_to_fp16)[name = string("aw_chunk_887_cast_fp16")]; fp16 var_9461_to_fp16 = const()[name = string("op_9461_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_889_cast_fp16, y = var_9461_to_fp16)[name = string("aw_chunk_889_cast_fp16")]; fp16 var_9463_to_fp16 = const()[name = string("op_9463_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_891_cast_fp16, y = var_9463_to_fp16)[name = string("aw_chunk_891_cast_fp16")]; fp16 var_9465_to_fp16 = const()[name = string("op_9465_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_893_cast_fp16, y = var_9465_to_fp16)[name = string("aw_chunk_893_cast_fp16")]; fp16 var_9467_to_fp16 = const()[name = string("op_9467_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_895_cast_fp16, y = var_9467_to_fp16)[name = string("aw_chunk_895_cast_fp16")]; fp16 var_9469_to_fp16 = const()[name = string("op_9469_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_897_cast_fp16, y = var_9469_to_fp16)[name = string("aw_chunk_897_cast_fp16")]; fp16 var_9471_to_fp16 = const()[name = string("op_9471_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_899_cast_fp16, y = var_9471_to_fp16)[name = string("aw_chunk_899_cast_fp16")]; fp16 var_9473_to_fp16 = const()[name = string("op_9473_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_901_cast_fp16, y = var_9473_to_fp16)[name = string("aw_chunk_901_cast_fp16")]; fp16 var_9475_to_fp16 = const()[name = string("op_9475_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_903_cast_fp16, y = var_9475_to_fp16)[name = string("aw_chunk_903_cast_fp16")]; fp16 var_9477_to_fp16 = const()[name = string("op_9477_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_905_cast_fp16, y = var_9477_to_fp16)[name = string("aw_chunk_905_cast_fp16")]; fp16 var_9479_to_fp16 = const()[name = string("op_9479_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_907_cast_fp16, y = var_9479_to_fp16)[name = string("aw_chunk_907_cast_fp16")]; fp16 var_9481_to_fp16 = const()[name = string("op_9481_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_909_cast_fp16, y = var_9481_to_fp16)[name = string("aw_chunk_909_cast_fp16")]; fp16 var_9483_to_fp16 = const()[name = string("op_9483_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_911_cast_fp16, y = var_9483_to_fp16)[name = string("aw_chunk_911_cast_fp16")]; fp16 var_9485_to_fp16 = const()[name = string("op_9485_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_913_cast_fp16, y = var_9485_to_fp16)[name = string("aw_chunk_913_cast_fp16")]; fp16 var_9487_to_fp16 = const()[name = string("op_9487_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_915_cast_fp16, y = var_9487_to_fp16)[name = string("aw_chunk_915_cast_fp16")]; fp16 var_9489_to_fp16 = const()[name = string("op_9489_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_917_cast_fp16, y = var_9489_to_fp16)[name = string("aw_chunk_917_cast_fp16")]; fp16 var_9491_to_fp16 = const()[name = string("op_9491_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_919_cast_fp16, y = var_9491_to_fp16)[name = string("aw_chunk_919_cast_fp16")]; fp16 var_9493_to_fp16 = const()[name = string("op_9493_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_921_cast_fp16, y = var_9493_to_fp16)[name = string("aw_chunk_921_cast_fp16")]; fp16 var_9495_to_fp16 = const()[name = string("op_9495_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_923_cast_fp16, y = var_9495_to_fp16)[name = string("aw_chunk_923_cast_fp16")]; fp16 var_9497_to_fp16 = const()[name = string("op_9497_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_925_cast_fp16, y = var_9497_to_fp16)[name = string("aw_chunk_925_cast_fp16")]; fp16 var_9499_to_fp16 = const()[name = string("op_9499_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_927_cast_fp16, y = var_9499_to_fp16)[name = string("aw_chunk_927_cast_fp16")]; fp16 var_9501_to_fp16 = const()[name = string("op_9501_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_929_cast_fp16, y = var_9501_to_fp16)[name = string("aw_chunk_929_cast_fp16")]; fp16 var_9503_to_fp16 = const()[name = string("op_9503_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_931_cast_fp16, y = var_9503_to_fp16)[name = string("aw_chunk_931_cast_fp16")]; fp16 var_9505_to_fp16 = const()[name = string("op_9505_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_933_cast_fp16, y = var_9505_to_fp16)[name = string("aw_chunk_933_cast_fp16")]; fp16 var_9507_to_fp16 = const()[name = string("op_9507_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_935_cast_fp16, y = var_9507_to_fp16)[name = string("aw_chunk_935_cast_fp16")]; fp16 var_9509_to_fp16 = const()[name = string("op_9509_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_937_cast_fp16, y = var_9509_to_fp16)[name = string("aw_chunk_937_cast_fp16")]; fp16 var_9511_to_fp16 = const()[name = string("op_9511_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_939_cast_fp16, y = var_9511_to_fp16)[name = string("aw_chunk_939_cast_fp16")]; fp16 var_9513_to_fp16 = const()[name = string("op_9513_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_941_cast_fp16, y = var_9513_to_fp16)[name = string("aw_chunk_941_cast_fp16")]; fp16 var_9515_to_fp16 = const()[name = string("op_9515_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_943_cast_fp16, y = var_9515_to_fp16)[name = string("aw_chunk_943_cast_fp16")]; fp16 var_9517_to_fp16 = const()[name = string("op_9517_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_945_cast_fp16, y = var_9517_to_fp16)[name = string("aw_chunk_945_cast_fp16")]; fp16 var_9519_to_fp16 = const()[name = string("op_9519_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_947_cast_fp16, y = var_9519_to_fp16)[name = string("aw_chunk_947_cast_fp16")]; fp16 var_9521_to_fp16 = const()[name = string("op_9521_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_949_cast_fp16, y = var_9521_to_fp16)[name = string("aw_chunk_949_cast_fp16")]; fp16 var_9523_to_fp16 = const()[name = string("op_9523_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_951_cast_fp16, y = var_9523_to_fp16)[name = string("aw_chunk_951_cast_fp16")]; fp16 var_9525_to_fp16 = const()[name = string("op_9525_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_953_cast_fp16, y = var_9525_to_fp16)[name = string("aw_chunk_953_cast_fp16")]; fp16 var_9527_to_fp16 = const()[name = string("op_9527_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_955_cast_fp16, y = var_9527_to_fp16)[name = string("aw_chunk_955_cast_fp16")]; fp16 var_9529_to_fp16 = const()[name = string("op_9529_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_957_cast_fp16, y = var_9529_to_fp16)[name = string("aw_chunk_957_cast_fp16")]; fp16 var_9531_to_fp16 = const()[name = string("op_9531_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_959_cast_fp16, y = var_9531_to_fp16)[name = string("aw_chunk_959_cast_fp16")]; tensor var_9533_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_865_cast_fp16)[name = string("op_9533_cast_fp16")]; tensor var_9534_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_867_cast_fp16)[name = string("op_9534_cast_fp16")]; tensor var_9535_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_869_cast_fp16)[name = string("op_9535_cast_fp16")]; tensor var_9536_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_871_cast_fp16)[name = string("op_9536_cast_fp16")]; tensor var_9537_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_873_cast_fp16)[name = string("op_9537_cast_fp16")]; tensor var_9538_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_875_cast_fp16)[name = string("op_9538_cast_fp16")]; tensor var_9539_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_877_cast_fp16)[name = string("op_9539_cast_fp16")]; tensor var_9540_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_879_cast_fp16)[name = string("op_9540_cast_fp16")]; tensor var_9541_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_881_cast_fp16)[name = string("op_9541_cast_fp16")]; tensor var_9542_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_883_cast_fp16)[name = string("op_9542_cast_fp16")]; tensor var_9543_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_885_cast_fp16)[name = string("op_9543_cast_fp16")]; tensor var_9544_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_887_cast_fp16)[name = string("op_9544_cast_fp16")]; tensor var_9545_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_889_cast_fp16)[name = string("op_9545_cast_fp16")]; tensor var_9546_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_891_cast_fp16)[name = string("op_9546_cast_fp16")]; tensor var_9547_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_893_cast_fp16)[name = string("op_9547_cast_fp16")]; tensor var_9548_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_895_cast_fp16)[name = string("op_9548_cast_fp16")]; tensor var_9549_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_897_cast_fp16)[name = string("op_9549_cast_fp16")]; tensor var_9550_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_899_cast_fp16)[name = string("op_9550_cast_fp16")]; tensor var_9551_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_901_cast_fp16)[name = string("op_9551_cast_fp16")]; tensor var_9552_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_903_cast_fp16)[name = string("op_9552_cast_fp16")]; tensor var_9553_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_905_cast_fp16)[name = string("op_9553_cast_fp16")]; tensor var_9554_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_907_cast_fp16)[name = string("op_9554_cast_fp16")]; tensor var_9555_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_909_cast_fp16)[name = string("op_9555_cast_fp16")]; tensor var_9556_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_911_cast_fp16)[name = string("op_9556_cast_fp16")]; tensor var_9557_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_913_cast_fp16)[name = string("op_9557_cast_fp16")]; tensor var_9558_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_915_cast_fp16)[name = string("op_9558_cast_fp16")]; tensor var_9559_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_917_cast_fp16)[name = string("op_9559_cast_fp16")]; tensor var_9560_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_919_cast_fp16)[name = string("op_9560_cast_fp16")]; tensor var_9561_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_921_cast_fp16)[name = string("op_9561_cast_fp16")]; tensor var_9562_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_923_cast_fp16)[name = string("op_9562_cast_fp16")]; tensor var_9563_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_925_cast_fp16)[name = string("op_9563_cast_fp16")]; tensor var_9564_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_927_cast_fp16)[name = string("op_9564_cast_fp16")]; tensor var_9565_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_929_cast_fp16)[name = string("op_9565_cast_fp16")]; tensor var_9566_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_931_cast_fp16)[name = string("op_9566_cast_fp16")]; tensor var_9567_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_933_cast_fp16)[name = string("op_9567_cast_fp16")]; tensor var_9568_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_935_cast_fp16)[name = string("op_9568_cast_fp16")]; tensor var_9569_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_937_cast_fp16)[name = string("op_9569_cast_fp16")]; tensor var_9570_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_939_cast_fp16)[name = string("op_9570_cast_fp16")]; tensor var_9571_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_941_cast_fp16)[name = string("op_9571_cast_fp16")]; tensor var_9572_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_943_cast_fp16)[name = string("op_9572_cast_fp16")]; tensor var_9573_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_945_cast_fp16)[name = string("op_9573_cast_fp16")]; tensor var_9574_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_947_cast_fp16)[name = string("op_9574_cast_fp16")]; tensor var_9575_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_949_cast_fp16)[name = string("op_9575_cast_fp16")]; tensor var_9576_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_951_cast_fp16)[name = string("op_9576_cast_fp16")]; tensor var_9577_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_953_cast_fp16)[name = string("op_9577_cast_fp16")]; tensor var_9578_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_955_cast_fp16)[name = string("op_9578_cast_fp16")]; tensor var_9579_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_957_cast_fp16)[name = string("op_9579_cast_fp16")]; tensor var_9580_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_959_cast_fp16)[name = string("op_9580_cast_fp16")]; string var_9582_equation_0 = const()[name = string("op_9582_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9582_cast_fp16 = einsum(equation = var_9582_equation_0, values = (var_9294_cast_fp16, var_9533_cast_fp16))[name = string("op_9582_cast_fp16")]; string var_9584_equation_0 = const()[name = string("op_9584_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9584_cast_fp16 = einsum(equation = var_9584_equation_0, values = (var_9294_cast_fp16, var_9534_cast_fp16))[name = string("op_9584_cast_fp16")]; string var_9586_equation_0 = const()[name = string("op_9586_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9586_cast_fp16 = einsum(equation = var_9586_equation_0, values = (var_9294_cast_fp16, var_9535_cast_fp16))[name = string("op_9586_cast_fp16")]; string var_9588_equation_0 = const()[name = string("op_9588_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9588_cast_fp16 = einsum(equation = var_9588_equation_0, values = (var_9294_cast_fp16, var_9536_cast_fp16))[name = string("op_9588_cast_fp16")]; string var_9590_equation_0 = const()[name = string("op_9590_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9590_cast_fp16 = einsum(equation = var_9590_equation_0, values = (var_9298_cast_fp16, var_9537_cast_fp16))[name = string("op_9590_cast_fp16")]; string var_9592_equation_0 = const()[name = string("op_9592_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9592_cast_fp16 = einsum(equation = var_9592_equation_0, values = (var_9298_cast_fp16, var_9538_cast_fp16))[name = string("op_9592_cast_fp16")]; string var_9594_equation_0 = const()[name = string("op_9594_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9594_cast_fp16 = einsum(equation = var_9594_equation_0, values = (var_9298_cast_fp16, var_9539_cast_fp16))[name = string("op_9594_cast_fp16")]; string var_9596_equation_0 = const()[name = string("op_9596_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9596_cast_fp16 = einsum(equation = var_9596_equation_0, values = (var_9298_cast_fp16, var_9540_cast_fp16))[name = string("op_9596_cast_fp16")]; string var_9598_equation_0 = const()[name = string("op_9598_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9598_cast_fp16 = einsum(equation = var_9598_equation_0, values = (var_9302_cast_fp16, var_9541_cast_fp16))[name = string("op_9598_cast_fp16")]; string var_9600_equation_0 = const()[name = string("op_9600_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9600_cast_fp16 = einsum(equation = var_9600_equation_0, values = (var_9302_cast_fp16, var_9542_cast_fp16))[name = string("op_9600_cast_fp16")]; string var_9602_equation_0 = const()[name = string("op_9602_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9602_cast_fp16 = einsum(equation = var_9602_equation_0, values = (var_9302_cast_fp16, var_9543_cast_fp16))[name = string("op_9602_cast_fp16")]; string var_9604_equation_0 = const()[name = string("op_9604_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9604_cast_fp16 = einsum(equation = var_9604_equation_0, values = (var_9302_cast_fp16, var_9544_cast_fp16))[name = string("op_9604_cast_fp16")]; string var_9606_equation_0 = const()[name = string("op_9606_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9606_cast_fp16 = einsum(equation = var_9606_equation_0, values = (var_9306_cast_fp16, var_9545_cast_fp16))[name = string("op_9606_cast_fp16")]; string var_9608_equation_0 = const()[name = string("op_9608_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9608_cast_fp16 = einsum(equation = var_9608_equation_0, values = (var_9306_cast_fp16, var_9546_cast_fp16))[name = string("op_9608_cast_fp16")]; string var_9610_equation_0 = const()[name = string("op_9610_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9610_cast_fp16 = einsum(equation = var_9610_equation_0, values = (var_9306_cast_fp16, var_9547_cast_fp16))[name = string("op_9610_cast_fp16")]; string var_9612_equation_0 = const()[name = string("op_9612_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9612_cast_fp16 = einsum(equation = var_9612_equation_0, values = (var_9306_cast_fp16, var_9548_cast_fp16))[name = string("op_9612_cast_fp16")]; string var_9614_equation_0 = const()[name = string("op_9614_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9614_cast_fp16 = einsum(equation = var_9614_equation_0, values = (var_9310_cast_fp16, var_9549_cast_fp16))[name = string("op_9614_cast_fp16")]; string var_9616_equation_0 = const()[name = string("op_9616_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9616_cast_fp16 = einsum(equation = var_9616_equation_0, values = (var_9310_cast_fp16, var_9550_cast_fp16))[name = string("op_9616_cast_fp16")]; string var_9618_equation_0 = const()[name = string("op_9618_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9618_cast_fp16 = einsum(equation = var_9618_equation_0, values = (var_9310_cast_fp16, var_9551_cast_fp16))[name = string("op_9618_cast_fp16")]; string var_9620_equation_0 = const()[name = string("op_9620_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9620_cast_fp16 = einsum(equation = var_9620_equation_0, values = (var_9310_cast_fp16, var_9552_cast_fp16))[name = string("op_9620_cast_fp16")]; string var_9622_equation_0 = const()[name = string("op_9622_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9622_cast_fp16 = einsum(equation = var_9622_equation_0, values = (var_9314_cast_fp16, var_9553_cast_fp16))[name = string("op_9622_cast_fp16")]; string var_9624_equation_0 = const()[name = string("op_9624_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9624_cast_fp16 = einsum(equation = var_9624_equation_0, values = (var_9314_cast_fp16, var_9554_cast_fp16))[name = string("op_9624_cast_fp16")]; string var_9626_equation_0 = const()[name = string("op_9626_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9626_cast_fp16 = einsum(equation = var_9626_equation_0, values = (var_9314_cast_fp16, var_9555_cast_fp16))[name = string("op_9626_cast_fp16")]; string var_9628_equation_0 = const()[name = string("op_9628_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9628_cast_fp16 = einsum(equation = var_9628_equation_0, values = (var_9314_cast_fp16, var_9556_cast_fp16))[name = string("op_9628_cast_fp16")]; string var_9630_equation_0 = const()[name = string("op_9630_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9630_cast_fp16 = einsum(equation = var_9630_equation_0, values = (var_9318_cast_fp16, var_9557_cast_fp16))[name = string("op_9630_cast_fp16")]; string var_9632_equation_0 = const()[name = string("op_9632_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9632_cast_fp16 = einsum(equation = var_9632_equation_0, values = (var_9318_cast_fp16, var_9558_cast_fp16))[name = string("op_9632_cast_fp16")]; string var_9634_equation_0 = const()[name = string("op_9634_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9634_cast_fp16 = einsum(equation = var_9634_equation_0, values = (var_9318_cast_fp16, var_9559_cast_fp16))[name = string("op_9634_cast_fp16")]; string var_9636_equation_0 = const()[name = string("op_9636_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9636_cast_fp16 = einsum(equation = var_9636_equation_0, values = (var_9318_cast_fp16, var_9560_cast_fp16))[name = string("op_9636_cast_fp16")]; string var_9638_equation_0 = const()[name = string("op_9638_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9638_cast_fp16 = einsum(equation = var_9638_equation_0, values = (var_9322_cast_fp16, var_9561_cast_fp16))[name = string("op_9638_cast_fp16")]; string var_9640_equation_0 = const()[name = string("op_9640_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9640_cast_fp16 = einsum(equation = var_9640_equation_0, values = (var_9322_cast_fp16, var_9562_cast_fp16))[name = string("op_9640_cast_fp16")]; string var_9642_equation_0 = const()[name = string("op_9642_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9642_cast_fp16 = einsum(equation = var_9642_equation_0, values = (var_9322_cast_fp16, var_9563_cast_fp16))[name = string("op_9642_cast_fp16")]; string var_9644_equation_0 = const()[name = string("op_9644_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9644_cast_fp16 = einsum(equation = var_9644_equation_0, values = (var_9322_cast_fp16, var_9564_cast_fp16))[name = string("op_9644_cast_fp16")]; string var_9646_equation_0 = const()[name = string("op_9646_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9646_cast_fp16 = einsum(equation = var_9646_equation_0, values = (var_9326_cast_fp16, var_9565_cast_fp16))[name = string("op_9646_cast_fp16")]; string var_9648_equation_0 = const()[name = string("op_9648_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9648_cast_fp16 = einsum(equation = var_9648_equation_0, values = (var_9326_cast_fp16, var_9566_cast_fp16))[name = string("op_9648_cast_fp16")]; string var_9650_equation_0 = const()[name = string("op_9650_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9650_cast_fp16 = einsum(equation = var_9650_equation_0, values = (var_9326_cast_fp16, var_9567_cast_fp16))[name = string("op_9650_cast_fp16")]; string var_9652_equation_0 = const()[name = string("op_9652_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9652_cast_fp16 = einsum(equation = var_9652_equation_0, values = (var_9326_cast_fp16, var_9568_cast_fp16))[name = string("op_9652_cast_fp16")]; string var_9654_equation_0 = const()[name = string("op_9654_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9654_cast_fp16 = einsum(equation = var_9654_equation_0, values = (var_9330_cast_fp16, var_9569_cast_fp16))[name = string("op_9654_cast_fp16")]; string var_9656_equation_0 = const()[name = string("op_9656_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9656_cast_fp16 = einsum(equation = var_9656_equation_0, values = (var_9330_cast_fp16, var_9570_cast_fp16))[name = string("op_9656_cast_fp16")]; string var_9658_equation_0 = const()[name = string("op_9658_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9658_cast_fp16 = einsum(equation = var_9658_equation_0, values = (var_9330_cast_fp16, var_9571_cast_fp16))[name = string("op_9658_cast_fp16")]; string var_9660_equation_0 = const()[name = string("op_9660_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9660_cast_fp16 = einsum(equation = var_9660_equation_0, values = (var_9330_cast_fp16, var_9572_cast_fp16))[name = string("op_9660_cast_fp16")]; string var_9662_equation_0 = const()[name = string("op_9662_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9662_cast_fp16 = einsum(equation = var_9662_equation_0, values = (var_9334_cast_fp16, var_9573_cast_fp16))[name = string("op_9662_cast_fp16")]; string var_9664_equation_0 = const()[name = string("op_9664_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9664_cast_fp16 = einsum(equation = var_9664_equation_0, values = (var_9334_cast_fp16, var_9574_cast_fp16))[name = string("op_9664_cast_fp16")]; string var_9666_equation_0 = const()[name = string("op_9666_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9666_cast_fp16 = einsum(equation = var_9666_equation_0, values = (var_9334_cast_fp16, var_9575_cast_fp16))[name = string("op_9666_cast_fp16")]; string var_9668_equation_0 = const()[name = string("op_9668_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9668_cast_fp16 = einsum(equation = var_9668_equation_0, values = (var_9334_cast_fp16, var_9576_cast_fp16))[name = string("op_9668_cast_fp16")]; string var_9670_equation_0 = const()[name = string("op_9670_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9670_cast_fp16 = einsum(equation = var_9670_equation_0, values = (var_9338_cast_fp16, var_9577_cast_fp16))[name = string("op_9670_cast_fp16")]; string var_9672_equation_0 = const()[name = string("op_9672_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9672_cast_fp16 = einsum(equation = var_9672_equation_0, values = (var_9338_cast_fp16, var_9578_cast_fp16))[name = string("op_9672_cast_fp16")]; string var_9674_equation_0 = const()[name = string("op_9674_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9674_cast_fp16 = einsum(equation = var_9674_equation_0, values = (var_9338_cast_fp16, var_9579_cast_fp16))[name = string("op_9674_cast_fp16")]; string var_9676_equation_0 = const()[name = string("op_9676_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_9676_cast_fp16 = einsum(equation = var_9676_equation_0, values = (var_9338_cast_fp16, var_9580_cast_fp16))[name = string("op_9676_cast_fp16")]; bool var_9678_interleave_0 = const()[name = string("op_9678_interleave_0"), val = bool(false)]; tensor var_9678_cast_fp16 = concat(axis = var_8789, interleave = var_9678_interleave_0, values = (var_9582_cast_fp16, var_9584_cast_fp16, var_9586_cast_fp16, var_9588_cast_fp16))[name = string("op_9678_cast_fp16")]; bool var_9680_interleave_0 = const()[name = string("op_9680_interleave_0"), val = bool(false)]; tensor var_9680_cast_fp16 = concat(axis = var_8789, interleave = var_9680_interleave_0, values = (var_9590_cast_fp16, var_9592_cast_fp16, var_9594_cast_fp16, var_9596_cast_fp16))[name = string("op_9680_cast_fp16")]; bool var_9682_interleave_0 = const()[name = string("op_9682_interleave_0"), val = bool(false)]; tensor var_9682_cast_fp16 = concat(axis = var_8789, interleave = var_9682_interleave_0, values = (var_9598_cast_fp16, var_9600_cast_fp16, var_9602_cast_fp16, var_9604_cast_fp16))[name = string("op_9682_cast_fp16")]; bool var_9684_interleave_0 = const()[name = string("op_9684_interleave_0"), val = bool(false)]; tensor var_9684_cast_fp16 = concat(axis = var_8789, interleave = var_9684_interleave_0, values = (var_9606_cast_fp16, var_9608_cast_fp16, var_9610_cast_fp16, var_9612_cast_fp16))[name = string("op_9684_cast_fp16")]; bool var_9686_interleave_0 = const()[name = string("op_9686_interleave_0"), val = bool(false)]; tensor var_9686_cast_fp16 = concat(axis = var_8789, interleave = var_9686_interleave_0, values = (var_9614_cast_fp16, var_9616_cast_fp16, var_9618_cast_fp16, var_9620_cast_fp16))[name = string("op_9686_cast_fp16")]; bool var_9688_interleave_0 = const()[name = string("op_9688_interleave_0"), val = bool(false)]; tensor var_9688_cast_fp16 = concat(axis = var_8789, interleave = var_9688_interleave_0, values = (var_9622_cast_fp16, var_9624_cast_fp16, var_9626_cast_fp16, var_9628_cast_fp16))[name = string("op_9688_cast_fp16")]; bool var_9690_interleave_0 = const()[name = string("op_9690_interleave_0"), val = bool(false)]; tensor var_9690_cast_fp16 = concat(axis = var_8789, interleave = var_9690_interleave_0, values = (var_9630_cast_fp16, var_9632_cast_fp16, var_9634_cast_fp16, var_9636_cast_fp16))[name = string("op_9690_cast_fp16")]; bool var_9692_interleave_0 = const()[name = string("op_9692_interleave_0"), val = bool(false)]; tensor var_9692_cast_fp16 = concat(axis = var_8789, interleave = var_9692_interleave_0, values = (var_9638_cast_fp16, var_9640_cast_fp16, var_9642_cast_fp16, var_9644_cast_fp16))[name = string("op_9692_cast_fp16")]; bool var_9694_interleave_0 = const()[name = string("op_9694_interleave_0"), val = bool(false)]; tensor var_9694_cast_fp16 = concat(axis = var_8789, interleave = var_9694_interleave_0, values = (var_9646_cast_fp16, var_9648_cast_fp16, var_9650_cast_fp16, var_9652_cast_fp16))[name = string("op_9694_cast_fp16")]; bool var_9696_interleave_0 = const()[name = string("op_9696_interleave_0"), val = bool(false)]; tensor var_9696_cast_fp16 = concat(axis = var_8789, interleave = var_9696_interleave_0, values = (var_9654_cast_fp16, var_9656_cast_fp16, var_9658_cast_fp16, var_9660_cast_fp16))[name = string("op_9696_cast_fp16")]; bool var_9698_interleave_0 = const()[name = string("op_9698_interleave_0"), val = bool(false)]; tensor var_9698_cast_fp16 = concat(axis = var_8789, interleave = var_9698_interleave_0, values = (var_9662_cast_fp16, var_9664_cast_fp16, var_9666_cast_fp16, var_9668_cast_fp16))[name = string("op_9698_cast_fp16")]; bool var_9700_interleave_0 = const()[name = string("op_9700_interleave_0"), val = bool(false)]; tensor var_9700_cast_fp16 = concat(axis = var_8789, interleave = var_9700_interleave_0, values = (var_9670_cast_fp16, var_9672_cast_fp16, var_9674_cast_fp16, var_9676_cast_fp16))[name = string("op_9700_cast_fp16")]; bool input_73_interleave_0 = const()[name = string("input_73_interleave_0"), val = bool(false)]; tensor input_73_cast_fp16 = concat(axis = var_8806, interleave = input_73_interleave_0, values = (var_9678_cast_fp16, var_9680_cast_fp16, var_9682_cast_fp16, var_9684_cast_fp16, var_9686_cast_fp16, var_9688_cast_fp16, var_9690_cast_fp16, var_9692_cast_fp16, var_9694_cast_fp16, var_9696_cast_fp16, var_9698_cast_fp16, var_9700_cast_fp16))[name = string("input_73_cast_fp16")]; string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137340288)))]; tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138520000)))]; tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; fp16 var_9719_to_fp16 = const()[name = string("op_9719_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_9719_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138521600)))]; tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138523200)))]; fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138524800)))]; tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143243456)))]; tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143249664)))]; tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147968320)))]; tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; int32 var_9748 = const()[name = string("op_9748"), val = int32(3)]; int32 var_9765 = const()[name = string("op_9765"), val = int32(1)]; tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; fp16 var_9782_to_fp16 = const()[name = string("op_9782_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_9782_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147969920)))]; tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147971520)))]; fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147973120)))]; tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149152832)))]; tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149154432)))]; tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334144)))]; tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151513856)))]; tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; tensor var_9820_begin_0 = const()[name = string("op_9820_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9820_end_0 = const()[name = string("op_9820_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9820_end_mask_0 = const()[name = string("op_9820_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9820_cast_fp16 = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = query_21_cast_fp16)[name = string("op_9820_cast_fp16")]; tensor var_9824_begin_0 = const()[name = string("op_9824_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9824_end_0 = const()[name = string("op_9824_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_9824_end_mask_0 = const()[name = string("op_9824_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9824_cast_fp16 = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = query_21_cast_fp16)[name = string("op_9824_cast_fp16")]; tensor var_9828_begin_0 = const()[name = string("op_9828_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9828_end_0 = const()[name = string("op_9828_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_9828_end_mask_0 = const()[name = string("op_9828_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9828_cast_fp16 = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = query_21_cast_fp16)[name = string("op_9828_cast_fp16")]; tensor var_9832_begin_0 = const()[name = string("op_9832_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9832_end_0 = const()[name = string("op_9832_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_9832_end_mask_0 = const()[name = string("op_9832_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9832_cast_fp16 = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = query_21_cast_fp16)[name = string("op_9832_cast_fp16")]; tensor var_9836_begin_0 = const()[name = string("op_9836_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9836_end_0 = const()[name = string("op_9836_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_9836_end_mask_0 = const()[name = string("op_9836_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9836_cast_fp16 = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = query_21_cast_fp16)[name = string("op_9836_cast_fp16")]; tensor var_9840_begin_0 = const()[name = string("op_9840_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9840_end_0 = const()[name = string("op_9840_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_9840_end_mask_0 = const()[name = string("op_9840_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9840_cast_fp16 = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = query_21_cast_fp16)[name = string("op_9840_cast_fp16")]; tensor var_9844_begin_0 = const()[name = string("op_9844_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9844_end_0 = const()[name = string("op_9844_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_9844_end_mask_0 = const()[name = string("op_9844_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9844_cast_fp16 = slice_by_index(begin = var_9844_begin_0, end = var_9844_end_0, end_mask = var_9844_end_mask_0, x = query_21_cast_fp16)[name = string("op_9844_cast_fp16")]; tensor var_9848_begin_0 = const()[name = string("op_9848_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9848_end_0 = const()[name = string("op_9848_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_9848_end_mask_0 = const()[name = string("op_9848_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9848_cast_fp16 = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = query_21_cast_fp16)[name = string("op_9848_cast_fp16")]; tensor var_9852_begin_0 = const()[name = string("op_9852_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9852_end_0 = const()[name = string("op_9852_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_9852_end_mask_0 = const()[name = string("op_9852_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9852_cast_fp16 = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = query_21_cast_fp16)[name = string("op_9852_cast_fp16")]; tensor var_9856_begin_0 = const()[name = string("op_9856_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9856_end_0 = const()[name = string("op_9856_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_9856_end_mask_0 = const()[name = string("op_9856_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9856_cast_fp16 = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = query_21_cast_fp16)[name = string("op_9856_cast_fp16")]; tensor var_9860_begin_0 = const()[name = string("op_9860_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_9860_end_0 = const()[name = string("op_9860_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_9860_end_mask_0 = const()[name = string("op_9860_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9860_cast_fp16 = slice_by_index(begin = var_9860_begin_0, end = var_9860_end_0, end_mask = var_9860_end_mask_0, x = query_21_cast_fp16)[name = string("op_9860_cast_fp16")]; tensor var_9864_begin_0 = const()[name = string("op_9864_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_9864_end_0 = const()[name = string("op_9864_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_9864_end_mask_0 = const()[name = string("op_9864_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9864_cast_fp16 = slice_by_index(begin = var_9864_begin_0, end = var_9864_end_0, end_mask = var_9864_end_mask_0, x = query_21_cast_fp16)[name = string("op_9864_cast_fp16")]; tensor var_9873_begin_0 = const()[name = string("op_9873_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9873_end_0 = const()[name = string("op_9873_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9873_end_mask_0 = const()[name = string("op_9873_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9873_cast_fp16 = slice_by_index(begin = var_9873_begin_0, end = var_9873_end_0, end_mask = var_9873_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9873_cast_fp16")]; tensor var_9880_begin_0 = const()[name = string("op_9880_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9880_end_0 = const()[name = string("op_9880_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9880_end_mask_0 = const()[name = string("op_9880_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9880_cast_fp16 = slice_by_index(begin = var_9880_begin_0, end = var_9880_end_0, end_mask = var_9880_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9880_cast_fp16")]; tensor var_9887_begin_0 = const()[name = string("op_9887_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9887_end_0 = const()[name = string("op_9887_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9887_end_mask_0 = const()[name = string("op_9887_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9887_cast_fp16 = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9887_cast_fp16")]; tensor var_9894_begin_0 = const()[name = string("op_9894_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9894_end_0 = const()[name = string("op_9894_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9894_end_mask_0 = const()[name = string("op_9894_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9894_cast_fp16 = slice_by_index(begin = var_9894_begin_0, end = var_9894_end_0, end_mask = var_9894_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9894_cast_fp16")]; tensor var_9901_begin_0 = const()[name = string("op_9901_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9901_end_0 = const()[name = string("op_9901_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9901_end_mask_0 = const()[name = string("op_9901_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9901_cast_fp16 = slice_by_index(begin = var_9901_begin_0, end = var_9901_end_0, end_mask = var_9901_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9901_cast_fp16")]; tensor var_9908_begin_0 = const()[name = string("op_9908_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9908_end_0 = const()[name = string("op_9908_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9908_end_mask_0 = const()[name = string("op_9908_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9908_cast_fp16 = slice_by_index(begin = var_9908_begin_0, end = var_9908_end_0, end_mask = var_9908_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9908_cast_fp16")]; tensor var_9915_begin_0 = const()[name = string("op_9915_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9915_end_0 = const()[name = string("op_9915_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9915_end_mask_0 = const()[name = string("op_9915_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9915_cast_fp16 = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9915_cast_fp16")]; tensor var_9922_begin_0 = const()[name = string("op_9922_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9922_end_0 = const()[name = string("op_9922_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9922_end_mask_0 = const()[name = string("op_9922_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9922_cast_fp16 = slice_by_index(begin = var_9922_begin_0, end = var_9922_end_0, end_mask = var_9922_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9922_cast_fp16")]; tensor var_9929_begin_0 = const()[name = string("op_9929_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9929_end_0 = const()[name = string("op_9929_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9929_end_mask_0 = const()[name = string("op_9929_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9929_cast_fp16 = slice_by_index(begin = var_9929_begin_0, end = var_9929_end_0, end_mask = var_9929_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9929_cast_fp16")]; tensor var_9936_begin_0 = const()[name = string("op_9936_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9936_end_0 = const()[name = string("op_9936_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9936_end_mask_0 = const()[name = string("op_9936_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9936_cast_fp16 = slice_by_index(begin = var_9936_begin_0, end = var_9936_end_0, end_mask = var_9936_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9936_cast_fp16")]; tensor var_9943_begin_0 = const()[name = string("op_9943_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9943_end_0 = const()[name = string("op_9943_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9943_end_mask_0 = const()[name = string("op_9943_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9943_cast_fp16 = slice_by_index(begin = var_9943_begin_0, end = var_9943_end_0, end_mask = var_9943_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9943_cast_fp16")]; tensor var_9950_begin_0 = const()[name = string("op_9950_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9950_end_0 = const()[name = string("op_9950_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9950_end_mask_0 = const()[name = string("op_9950_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9950_cast_fp16 = slice_by_index(begin = var_9950_begin_0, end = var_9950_end_0, end_mask = var_9950_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9950_cast_fp16")]; tensor var_9957_begin_0 = const()[name = string("op_9957_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9957_end_0 = const()[name = string("op_9957_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9957_end_mask_0 = const()[name = string("op_9957_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9957_cast_fp16 = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9957_cast_fp16")]; tensor var_9964_begin_0 = const()[name = string("op_9964_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9964_end_0 = const()[name = string("op_9964_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9964_end_mask_0 = const()[name = string("op_9964_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9964_cast_fp16 = slice_by_index(begin = var_9964_begin_0, end = var_9964_end_0, end_mask = var_9964_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9964_cast_fp16")]; tensor var_9971_begin_0 = const()[name = string("op_9971_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9971_end_0 = const()[name = string("op_9971_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9971_end_mask_0 = const()[name = string("op_9971_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9971_cast_fp16 = slice_by_index(begin = var_9971_begin_0, end = var_9971_end_0, end_mask = var_9971_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9971_cast_fp16")]; tensor var_9978_begin_0 = const()[name = string("op_9978_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9978_end_0 = const()[name = string("op_9978_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9978_end_mask_0 = const()[name = string("op_9978_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9978_cast_fp16 = slice_by_index(begin = var_9978_begin_0, end = var_9978_end_0, end_mask = var_9978_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9978_cast_fp16")]; tensor var_9985_begin_0 = const()[name = string("op_9985_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9985_end_0 = const()[name = string("op_9985_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9985_end_mask_0 = const()[name = string("op_9985_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9985_cast_fp16 = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9985_cast_fp16")]; tensor var_9992_begin_0 = const()[name = string("op_9992_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9992_end_0 = const()[name = string("op_9992_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9992_end_mask_0 = const()[name = string("op_9992_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9992_cast_fp16 = slice_by_index(begin = var_9992_begin_0, end = var_9992_end_0, end_mask = var_9992_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9992_cast_fp16")]; tensor var_9999_begin_0 = const()[name = string("op_9999_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9999_end_0 = const()[name = string("op_9999_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9999_end_mask_0 = const()[name = string("op_9999_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9999_cast_fp16 = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9999_cast_fp16")]; tensor var_10006_begin_0 = const()[name = string("op_10006_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10006_end_0 = const()[name = string("op_10006_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10006_end_mask_0 = const()[name = string("op_10006_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10006_cast_fp16 = slice_by_index(begin = var_10006_begin_0, end = var_10006_end_0, end_mask = var_10006_end_mask_0, x = var_9836_cast_fp16)[name = string("op_10006_cast_fp16")]; tensor var_10013_begin_0 = const()[name = string("op_10013_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10013_end_0 = const()[name = string("op_10013_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10013_end_mask_0 = const()[name = string("op_10013_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10013_cast_fp16 = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10013_cast_fp16")]; tensor var_10020_begin_0 = const()[name = string("op_10020_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10020_end_0 = const()[name = string("op_10020_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10020_end_mask_0 = const()[name = string("op_10020_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10020_cast_fp16 = slice_by_index(begin = var_10020_begin_0, end = var_10020_end_0, end_mask = var_10020_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10020_cast_fp16")]; tensor var_10027_begin_0 = const()[name = string("op_10027_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10027_end_0 = const()[name = string("op_10027_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10027_end_mask_0 = const()[name = string("op_10027_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10027_cast_fp16 = slice_by_index(begin = var_10027_begin_0, end = var_10027_end_0, end_mask = var_10027_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10027_cast_fp16")]; tensor var_10034_begin_0 = const()[name = string("op_10034_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10034_end_0 = const()[name = string("op_10034_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10034_end_mask_0 = const()[name = string("op_10034_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10034_cast_fp16")]; tensor var_10041_begin_0 = const()[name = string("op_10041_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10041_end_0 = const()[name = string("op_10041_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10041_end_mask_0 = const()[name = string("op_10041_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10041_cast_fp16 = slice_by_index(begin = var_10041_begin_0, end = var_10041_end_0, end_mask = var_10041_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10041_cast_fp16")]; tensor var_10048_begin_0 = const()[name = string("op_10048_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10048_end_0 = const()[name = string("op_10048_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10048_end_mask_0 = const()[name = string("op_10048_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10048_cast_fp16 = slice_by_index(begin = var_10048_begin_0, end = var_10048_end_0, end_mask = var_10048_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10048_cast_fp16")]; tensor var_10055_begin_0 = const()[name = string("op_10055_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10055_end_0 = const()[name = string("op_10055_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10055_end_mask_0 = const()[name = string("op_10055_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10055_cast_fp16 = slice_by_index(begin = var_10055_begin_0, end = var_10055_end_0, end_mask = var_10055_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10055_cast_fp16")]; tensor var_10062_begin_0 = const()[name = string("op_10062_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10062_end_0 = const()[name = string("op_10062_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10062_end_mask_0 = const()[name = string("op_10062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10062_cast_fp16")]; tensor var_10069_begin_0 = const()[name = string("op_10069_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10069_end_0 = const()[name = string("op_10069_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10069_end_mask_0 = const()[name = string("op_10069_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10069_cast_fp16 = slice_by_index(begin = var_10069_begin_0, end = var_10069_end_0, end_mask = var_10069_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10069_cast_fp16")]; tensor var_10076_begin_0 = const()[name = string("op_10076_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10076_end_0 = const()[name = string("op_10076_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10076_end_mask_0 = const()[name = string("op_10076_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10076_cast_fp16")]; tensor var_10083_begin_0 = const()[name = string("op_10083_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10083_end_0 = const()[name = string("op_10083_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10083_end_mask_0 = const()[name = string("op_10083_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10083_cast_fp16 = slice_by_index(begin = var_10083_begin_0, end = var_10083_end_0, end_mask = var_10083_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10083_cast_fp16")]; tensor var_10090_begin_0 = const()[name = string("op_10090_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10090_end_0 = const()[name = string("op_10090_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10090_end_mask_0 = const()[name = string("op_10090_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10090_cast_fp16 = slice_by_index(begin = var_10090_begin_0, end = var_10090_end_0, end_mask = var_10090_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10090_cast_fp16")]; tensor var_10097_begin_0 = const()[name = string("op_10097_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10097_end_0 = const()[name = string("op_10097_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10097_end_mask_0 = const()[name = string("op_10097_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10097_cast_fp16 = slice_by_index(begin = var_10097_begin_0, end = var_10097_end_0, end_mask = var_10097_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10097_cast_fp16")]; tensor var_10104_begin_0 = const()[name = string("op_10104_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10104_end_0 = const()[name = string("op_10104_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10104_end_mask_0 = const()[name = string("op_10104_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10104_cast_fp16")]; tensor var_10111_begin_0 = const()[name = string("op_10111_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10111_end_0 = const()[name = string("op_10111_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10111_end_mask_0 = const()[name = string("op_10111_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10111_cast_fp16 = slice_by_index(begin = var_10111_begin_0, end = var_10111_end_0, end_mask = var_10111_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10111_cast_fp16")]; tensor var_10118_begin_0 = const()[name = string("op_10118_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10118_end_0 = const()[name = string("op_10118_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10118_end_mask_0 = const()[name = string("op_10118_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10118_cast_fp16 = slice_by_index(begin = var_10118_begin_0, end = var_10118_end_0, end_mask = var_10118_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10118_cast_fp16")]; tensor var_10125_begin_0 = const()[name = string("op_10125_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10125_end_0 = const()[name = string("op_10125_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10125_end_mask_0 = const()[name = string("op_10125_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10125_cast_fp16 = slice_by_index(begin = var_10125_begin_0, end = var_10125_end_0, end_mask = var_10125_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10125_cast_fp16")]; tensor var_10132_begin_0 = const()[name = string("op_10132_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10132_end_0 = const()[name = string("op_10132_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10132_end_mask_0 = const()[name = string("op_10132_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10132_cast_fp16 = slice_by_index(begin = var_10132_begin_0, end = var_10132_end_0, end_mask = var_10132_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10132_cast_fp16")]; tensor var_10139_begin_0 = const()[name = string("op_10139_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10139_end_0 = const()[name = string("op_10139_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10139_end_mask_0 = const()[name = string("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10139_cast_fp16")]; tensor var_10146_begin_0 = const()[name = string("op_10146_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10146_end_0 = const()[name = string("op_10146_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10146_end_mask_0 = const()[name = string("op_10146_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10146_cast_fp16 = slice_by_index(begin = var_10146_begin_0, end = var_10146_end_0, end_mask = var_10146_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10146_cast_fp16")]; tensor var_10153_begin_0 = const()[name = string("op_10153_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10153_end_0 = const()[name = string("op_10153_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10153_end_mask_0 = const()[name = string("op_10153_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10153_cast_fp16 = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10153_cast_fp16")]; tensor var_10160_begin_0 = const()[name = string("op_10160_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10160_end_0 = const()[name = string("op_10160_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10160_end_mask_0 = const()[name = string("op_10160_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10160_cast_fp16 = slice_by_index(begin = var_10160_begin_0, end = var_10160_end_0, end_mask = var_10160_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10160_cast_fp16")]; tensor var_10167_begin_0 = const()[name = string("op_10167_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10167_end_0 = const()[name = string("op_10167_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10167_end_mask_0 = const()[name = string("op_10167_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10167_cast_fp16 = slice_by_index(begin = var_10167_begin_0, end = var_10167_end_0, end_mask = var_10167_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10167_cast_fp16")]; tensor var_10174_begin_0 = const()[name = string("op_10174_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10174_end_0 = const()[name = string("op_10174_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10174_end_mask_0 = const()[name = string("op_10174_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10174_cast_fp16 = slice_by_index(begin = var_10174_begin_0, end = var_10174_end_0, end_mask = var_10174_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10174_cast_fp16")]; tensor var_10181_begin_0 = const()[name = string("op_10181_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10181_end_0 = const()[name = string("op_10181_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10181_end_mask_0 = const()[name = string("op_10181_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10181_cast_fp16 = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10181_cast_fp16")]; tensor var_10188_begin_0 = const()[name = string("op_10188_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10188_end_0 = const()[name = string("op_10188_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10188_end_mask_0 = const()[name = string("op_10188_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10188_cast_fp16 = slice_by_index(begin = var_10188_begin_0, end = var_10188_end_0, end_mask = var_10188_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10188_cast_fp16")]; tensor var_10195_begin_0 = const()[name = string("op_10195_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10195_end_0 = const()[name = string("op_10195_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10195_end_mask_0 = const()[name = string("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10195_cast_fp16")]; tensor var_10202_begin_0 = const()[name = string("op_10202_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10202_end_0 = const()[name = string("op_10202_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10202_end_mask_0 = const()[name = string("op_10202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10202_cast_fp16 = slice_by_index(begin = var_10202_begin_0, end = var_10202_end_0, end_mask = var_10202_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10202_cast_fp16")]; tensor k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10207_begin_0 = const()[name = string("op_10207_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10207_end_0 = const()[name = string("op_10207_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_10207_end_mask_0 = const()[name = string("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = string("transpose_1")]; tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = k_21_cast_fp16)[name = string("op_10207_cast_fp16")]; tensor var_10211_begin_0 = const()[name = string("op_10211_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10211_end_0 = const()[name = string("op_10211_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_10211_end_mask_0 = const()[name = string("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = k_21_cast_fp16)[name = string("op_10211_cast_fp16")]; tensor var_10215_begin_0 = const()[name = string("op_10215_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10215_end_0 = const()[name = string("op_10215_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_10215_end_mask_0 = const()[name = string("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = k_21_cast_fp16)[name = string("op_10215_cast_fp16")]; tensor var_10219_begin_0 = const()[name = string("op_10219_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10219_end_0 = const()[name = string("op_10219_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_10219_end_mask_0 = const()[name = string("op_10219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10219_cast_fp16 = slice_by_index(begin = var_10219_begin_0, end = var_10219_end_0, end_mask = var_10219_end_mask_0, x = k_21_cast_fp16)[name = string("op_10219_cast_fp16")]; tensor var_10223_begin_0 = const()[name = string("op_10223_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10223_end_0 = const()[name = string("op_10223_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_10223_end_mask_0 = const()[name = string("op_10223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10223_cast_fp16 = slice_by_index(begin = var_10223_begin_0, end = var_10223_end_0, end_mask = var_10223_end_mask_0, x = k_21_cast_fp16)[name = string("op_10223_cast_fp16")]; tensor var_10227_begin_0 = const()[name = string("op_10227_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_10227_end_0 = const()[name = string("op_10227_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_10227_end_mask_0 = const()[name = string("op_10227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10227_cast_fp16 = slice_by_index(begin = var_10227_begin_0, end = var_10227_end_0, end_mask = var_10227_end_mask_0, x = k_21_cast_fp16)[name = string("op_10227_cast_fp16")]; tensor var_10231_begin_0 = const()[name = string("op_10231_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_10231_end_0 = const()[name = string("op_10231_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_10231_end_mask_0 = const()[name = string("op_10231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10231_cast_fp16 = slice_by_index(begin = var_10231_begin_0, end = var_10231_end_0, end_mask = var_10231_end_mask_0, x = k_21_cast_fp16)[name = string("op_10231_cast_fp16")]; tensor var_10235_begin_0 = const()[name = string("op_10235_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_10235_end_0 = const()[name = string("op_10235_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_10235_end_mask_0 = const()[name = string("op_10235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10235_cast_fp16 = slice_by_index(begin = var_10235_begin_0, end = var_10235_end_0, end_mask = var_10235_end_mask_0, x = k_21_cast_fp16)[name = string("op_10235_cast_fp16")]; tensor var_10239_begin_0 = const()[name = string("op_10239_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10239_end_0 = const()[name = string("op_10239_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_10239_end_mask_0 = const()[name = string("op_10239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10239_cast_fp16 = slice_by_index(begin = var_10239_begin_0, end = var_10239_end_0, end_mask = var_10239_end_mask_0, x = k_21_cast_fp16)[name = string("op_10239_cast_fp16")]; tensor var_10243_begin_0 = const()[name = string("op_10243_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_10243_end_0 = const()[name = string("op_10243_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_10243_end_mask_0 = const()[name = string("op_10243_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10243_cast_fp16 = slice_by_index(begin = var_10243_begin_0, end = var_10243_end_0, end_mask = var_10243_end_mask_0, x = k_21_cast_fp16)[name = string("op_10243_cast_fp16")]; tensor var_10247_begin_0 = const()[name = string("op_10247_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_10247_end_0 = const()[name = string("op_10247_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_10247_end_mask_0 = const()[name = string("op_10247_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10247_cast_fp16 = slice_by_index(begin = var_10247_begin_0, end = var_10247_end_0, end_mask = var_10247_end_mask_0, x = k_21_cast_fp16)[name = string("op_10247_cast_fp16")]; tensor var_10251_begin_0 = const()[name = string("op_10251_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_10251_end_0 = const()[name = string("op_10251_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_10251_end_mask_0 = const()[name = string("op_10251_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10251_cast_fp16 = slice_by_index(begin = var_10251_begin_0, end = var_10251_end_0, end_mask = var_10251_end_mask_0, x = k_21_cast_fp16)[name = string("op_10251_cast_fp16")]; tensor var_10253_begin_0 = const()[name = string("op_10253_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10253_end_0 = const()[name = string("op_10253_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10253_end_mask_0 = const()[name = string("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = value_21_cast_fp16)[name = string("op_10253_cast_fp16")]; tensor var_10257_begin_0 = const()[name = string("op_10257_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10257_end_0 = const()[name = string("op_10257_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_10257_end_mask_0 = const()[name = string("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = value_21_cast_fp16)[name = string("op_10257_cast_fp16")]; tensor var_10261_begin_0 = const()[name = string("op_10261_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10261_end_0 = const()[name = string("op_10261_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_10261_end_mask_0 = const()[name = string("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = value_21_cast_fp16)[name = string("op_10261_cast_fp16")]; tensor var_10265_begin_0 = const()[name = string("op_10265_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10265_end_0 = const()[name = string("op_10265_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_10265_end_mask_0 = const()[name = string("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = value_21_cast_fp16)[name = string("op_10265_cast_fp16")]; tensor var_10269_begin_0 = const()[name = string("op_10269_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10269_end_0 = const()[name = string("op_10269_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_10269_end_mask_0 = const()[name = string("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = value_21_cast_fp16)[name = string("op_10269_cast_fp16")]; tensor var_10273_begin_0 = const()[name = string("op_10273_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10273_end_0 = const()[name = string("op_10273_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_10273_end_mask_0 = const()[name = string("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = value_21_cast_fp16)[name = string("op_10273_cast_fp16")]; tensor var_10277_begin_0 = const()[name = string("op_10277_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10277_end_0 = const()[name = string("op_10277_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_10277_end_mask_0 = const()[name = string("op_10277_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10277_cast_fp16 = slice_by_index(begin = var_10277_begin_0, end = var_10277_end_0, end_mask = var_10277_end_mask_0, x = value_21_cast_fp16)[name = string("op_10277_cast_fp16")]; tensor var_10281_begin_0 = const()[name = string("op_10281_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10281_end_0 = const()[name = string("op_10281_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_10281_end_mask_0 = const()[name = string("op_10281_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10281_cast_fp16 = slice_by_index(begin = var_10281_begin_0, end = var_10281_end_0, end_mask = var_10281_end_mask_0, x = value_21_cast_fp16)[name = string("op_10281_cast_fp16")]; tensor var_10285_begin_0 = const()[name = string("op_10285_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10285_end_0 = const()[name = string("op_10285_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_10285_end_mask_0 = const()[name = string("op_10285_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10285_cast_fp16 = slice_by_index(begin = var_10285_begin_0, end = var_10285_end_0, end_mask = var_10285_end_mask_0, x = value_21_cast_fp16)[name = string("op_10285_cast_fp16")]; tensor var_10289_begin_0 = const()[name = string("op_10289_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10289_end_0 = const()[name = string("op_10289_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_10289_end_mask_0 = const()[name = string("op_10289_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10289_cast_fp16 = slice_by_index(begin = var_10289_begin_0, end = var_10289_end_0, end_mask = var_10289_end_mask_0, x = value_21_cast_fp16)[name = string("op_10289_cast_fp16")]; tensor var_10293_begin_0 = const()[name = string("op_10293_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_10293_end_0 = const()[name = string("op_10293_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_10293_end_mask_0 = const()[name = string("op_10293_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10293_cast_fp16 = slice_by_index(begin = var_10293_begin_0, end = var_10293_end_0, end_mask = var_10293_end_mask_0, x = value_21_cast_fp16)[name = string("op_10293_cast_fp16")]; tensor var_10297_begin_0 = const()[name = string("op_10297_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_10297_end_0 = const()[name = string("op_10297_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_10297_end_mask_0 = const()[name = string("op_10297_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10297_cast_fp16 = slice_by_index(begin = var_10297_begin_0, end = var_10297_end_0, end_mask = var_10297_end_mask_0, x = value_21_cast_fp16)[name = string("op_10297_cast_fp16")]; string _SplitHeadsQ__mh_w_961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_961_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_961_equation_0, values = (var_10207_cast_fp16, var_9873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_961_cast_fp16")]; string _SplitHeadsQ__mh_w_963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_963_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_963_equation_0, values = (var_10207_cast_fp16, var_9880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_963_cast_fp16")]; string _SplitHeadsQ__mh_w_965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_965_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_965_equation_0, values = (var_10207_cast_fp16, var_9887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_965_cast_fp16")]; string _SplitHeadsQ__mh_w_967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_967_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_967_equation_0, values = (var_10207_cast_fp16, var_9894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_967_cast_fp16")]; string _SplitHeadsQ__mh_w_969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_969_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_969_equation_0, values = (var_10211_cast_fp16, var_9901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_969_cast_fp16")]; string _SplitHeadsQ__mh_w_971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_971_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_971_equation_0, values = (var_10211_cast_fp16, var_9908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_971_cast_fp16")]; string _SplitHeadsQ__mh_w_973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_973_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_973_equation_0, values = (var_10211_cast_fp16, var_9915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_973_cast_fp16")]; string _SplitHeadsQ__mh_w_975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_975_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_975_equation_0, values = (var_10211_cast_fp16, var_9922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_975_cast_fp16")]; string _SplitHeadsQ__mh_w_977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_977_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_977_equation_0, values = (var_10215_cast_fp16, var_9929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_977_cast_fp16")]; string _SplitHeadsQ__mh_w_979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_979_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_979_equation_0, values = (var_10215_cast_fp16, var_9936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_979_cast_fp16")]; string _SplitHeadsQ__mh_w_981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_981_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_981_equation_0, values = (var_10215_cast_fp16, var_9943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_981_cast_fp16")]; string _SplitHeadsQ__mh_w_983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_983_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_983_equation_0, values = (var_10215_cast_fp16, var_9950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_983_cast_fp16")]; string _SplitHeadsQ__mh_w_985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_985_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_985_equation_0, values = (var_10219_cast_fp16, var_9957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_985_cast_fp16")]; string _SplitHeadsQ__mh_w_987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_987_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_987_equation_0, values = (var_10219_cast_fp16, var_9964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_987_cast_fp16")]; string _SplitHeadsQ__mh_w_989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_989_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_989_equation_0, values = (var_10219_cast_fp16, var_9971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_989_cast_fp16")]; string _SplitHeadsQ__mh_w_991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_991_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_991_equation_0, values = (var_10219_cast_fp16, var_9978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_991_cast_fp16")]; string _SplitHeadsQ__mh_w_993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_993_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_993_equation_0, values = (var_10223_cast_fp16, var_9985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_993_cast_fp16")]; string _SplitHeadsQ__mh_w_995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_995_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_995_equation_0, values = (var_10223_cast_fp16, var_9992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_995_cast_fp16")]; string _SplitHeadsQ__mh_w_997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_997_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_997_equation_0, values = (var_10223_cast_fp16, var_9999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_997_cast_fp16")]; string _SplitHeadsQ__mh_w_999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_999_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_999_equation_0, values = (var_10223_cast_fp16, var_10006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_999_cast_fp16")]; string _SplitHeadsQ__mh_w_1001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1001_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1001_equation_0, values = (var_10227_cast_fp16, var_10013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1001_cast_fp16")]; string _SplitHeadsQ__mh_w_1003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1003_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1003_equation_0, values = (var_10227_cast_fp16, var_10020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1003_cast_fp16")]; string _SplitHeadsQ__mh_w_1005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1005_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1005_equation_0, values = (var_10227_cast_fp16, var_10027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1005_cast_fp16")]; string _SplitHeadsQ__mh_w_1007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1007_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1007_equation_0, values = (var_10227_cast_fp16, var_10034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1007_cast_fp16")]; string _SplitHeadsQ__mh_w_1009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1009_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1009_equation_0, values = (var_10231_cast_fp16, var_10041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1009_cast_fp16")]; string _SplitHeadsQ__mh_w_1011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1011_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1011_equation_0, values = (var_10231_cast_fp16, var_10048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1011_cast_fp16")]; string _SplitHeadsQ__mh_w_1013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1013_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1013_equation_0, values = (var_10231_cast_fp16, var_10055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1013_cast_fp16")]; string _SplitHeadsQ__mh_w_1015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1015_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1015_equation_0, values = (var_10231_cast_fp16, var_10062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1015_cast_fp16")]; string _SplitHeadsQ__mh_w_1017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1017_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1017_equation_0, values = (var_10235_cast_fp16, var_10069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1017_cast_fp16")]; string _SplitHeadsQ__mh_w_1019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1019_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1019_equation_0, values = (var_10235_cast_fp16, var_10076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1019_cast_fp16")]; string _SplitHeadsQ__mh_w_1021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1021_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1021_equation_0, values = (var_10235_cast_fp16, var_10083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1021_cast_fp16")]; string _SplitHeadsQ__mh_w_1023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1023_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1023_equation_0, values = (var_10235_cast_fp16, var_10090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1023_cast_fp16")]; string _SplitHeadsQ__mh_w_1025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1025_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1025_equation_0, values = (var_10239_cast_fp16, var_10097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1025_cast_fp16")]; string _SplitHeadsQ__mh_w_1027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1027_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1027_equation_0, values = (var_10239_cast_fp16, var_10104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1027_cast_fp16")]; string _SplitHeadsQ__mh_w_1029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1029_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1029_equation_0, values = (var_10239_cast_fp16, var_10111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1029_cast_fp16")]; string _SplitHeadsQ__mh_w_1031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1031_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1031_equation_0, values = (var_10239_cast_fp16, var_10118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1031_cast_fp16")]; string _SplitHeadsQ__mh_w_1033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1033_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1033_equation_0, values = (var_10243_cast_fp16, var_10125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1033_cast_fp16")]; string _SplitHeadsQ__mh_w_1035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1035_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1035_equation_0, values = (var_10243_cast_fp16, var_10132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1035_cast_fp16")]; string _SplitHeadsQ__mh_w_1037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1037_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1037_equation_0, values = (var_10243_cast_fp16, var_10139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1037_cast_fp16")]; string _SplitHeadsQ__mh_w_1039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1039_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1039_equation_0, values = (var_10243_cast_fp16, var_10146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1039_cast_fp16")]; string _SplitHeadsQ__mh_w_1041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1041_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1041_equation_0, values = (var_10247_cast_fp16, var_10153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1041_cast_fp16")]; string _SplitHeadsQ__mh_w_1043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1043_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1043_equation_0, values = (var_10247_cast_fp16, var_10160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1043_cast_fp16")]; string _SplitHeadsQ__mh_w_1045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1045_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1045_equation_0, values = (var_10247_cast_fp16, var_10167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1045_cast_fp16")]; string _SplitHeadsQ__mh_w_1047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1047_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1047_equation_0, values = (var_10247_cast_fp16, var_10174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1047_cast_fp16")]; string _SplitHeadsQ__mh_w_1049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1049_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1049_equation_0, values = (var_10251_cast_fp16, var_10181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1049_cast_fp16")]; string _SplitHeadsQ__mh_w_1051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1051_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1051_equation_0, values = (var_10251_cast_fp16, var_10188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1051_cast_fp16")]; string _SplitHeadsQ__mh_w_1053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1053_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1053_equation_0, values = (var_10251_cast_fp16, var_10195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1053_cast_fp16")]; string _SplitHeadsQ__mh_w_1055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1055_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1055_equation_0, values = (var_10251_cast_fp16, var_10202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1055_cast_fp16")]; fp16 var_10396_to_fp16 = const()[name = string("op_10396_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_961_cast_fp16, y = var_10396_to_fp16)[name = string("aw_chunk_961_cast_fp16")]; fp16 var_10398_to_fp16 = const()[name = string("op_10398_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_963_cast_fp16, y = var_10398_to_fp16)[name = string("aw_chunk_963_cast_fp16")]; fp16 var_10400_to_fp16 = const()[name = string("op_10400_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_965_cast_fp16, y = var_10400_to_fp16)[name = string("aw_chunk_965_cast_fp16")]; fp16 var_10402_to_fp16 = const()[name = string("op_10402_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_967_cast_fp16, y = var_10402_to_fp16)[name = string("aw_chunk_967_cast_fp16")]; fp16 var_10404_to_fp16 = const()[name = string("op_10404_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_969_cast_fp16, y = var_10404_to_fp16)[name = string("aw_chunk_969_cast_fp16")]; fp16 var_10406_to_fp16 = const()[name = string("op_10406_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_971_cast_fp16, y = var_10406_to_fp16)[name = string("aw_chunk_971_cast_fp16")]; fp16 var_10408_to_fp16 = const()[name = string("op_10408_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_973_cast_fp16, y = var_10408_to_fp16)[name = string("aw_chunk_973_cast_fp16")]; fp16 var_10410_to_fp16 = const()[name = string("op_10410_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_975_cast_fp16, y = var_10410_to_fp16)[name = string("aw_chunk_975_cast_fp16")]; fp16 var_10412_to_fp16 = const()[name = string("op_10412_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_977_cast_fp16, y = var_10412_to_fp16)[name = string("aw_chunk_977_cast_fp16")]; fp16 var_10414_to_fp16 = const()[name = string("op_10414_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_979_cast_fp16, y = var_10414_to_fp16)[name = string("aw_chunk_979_cast_fp16")]; fp16 var_10416_to_fp16 = const()[name = string("op_10416_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_981_cast_fp16, y = var_10416_to_fp16)[name = string("aw_chunk_981_cast_fp16")]; fp16 var_10418_to_fp16 = const()[name = string("op_10418_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_983_cast_fp16, y = var_10418_to_fp16)[name = string("aw_chunk_983_cast_fp16")]; fp16 var_10420_to_fp16 = const()[name = string("op_10420_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_985_cast_fp16, y = var_10420_to_fp16)[name = string("aw_chunk_985_cast_fp16")]; fp16 var_10422_to_fp16 = const()[name = string("op_10422_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_987_cast_fp16, y = var_10422_to_fp16)[name = string("aw_chunk_987_cast_fp16")]; fp16 var_10424_to_fp16 = const()[name = string("op_10424_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_989_cast_fp16, y = var_10424_to_fp16)[name = string("aw_chunk_989_cast_fp16")]; fp16 var_10426_to_fp16 = const()[name = string("op_10426_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_991_cast_fp16, y = var_10426_to_fp16)[name = string("aw_chunk_991_cast_fp16")]; fp16 var_10428_to_fp16 = const()[name = string("op_10428_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_993_cast_fp16, y = var_10428_to_fp16)[name = string("aw_chunk_993_cast_fp16")]; fp16 var_10430_to_fp16 = const()[name = string("op_10430_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_995_cast_fp16, y = var_10430_to_fp16)[name = string("aw_chunk_995_cast_fp16")]; fp16 var_10432_to_fp16 = const()[name = string("op_10432_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_997_cast_fp16, y = var_10432_to_fp16)[name = string("aw_chunk_997_cast_fp16")]; fp16 var_10434_to_fp16 = const()[name = string("op_10434_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_999_cast_fp16, y = var_10434_to_fp16)[name = string("aw_chunk_999_cast_fp16")]; fp16 var_10436_to_fp16 = const()[name = string("op_10436_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1001_cast_fp16, y = var_10436_to_fp16)[name = string("aw_chunk_1001_cast_fp16")]; fp16 var_10438_to_fp16 = const()[name = string("op_10438_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1003_cast_fp16, y = var_10438_to_fp16)[name = string("aw_chunk_1003_cast_fp16")]; fp16 var_10440_to_fp16 = const()[name = string("op_10440_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1005_cast_fp16, y = var_10440_to_fp16)[name = string("aw_chunk_1005_cast_fp16")]; fp16 var_10442_to_fp16 = const()[name = string("op_10442_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1007_cast_fp16, y = var_10442_to_fp16)[name = string("aw_chunk_1007_cast_fp16")]; fp16 var_10444_to_fp16 = const()[name = string("op_10444_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1009_cast_fp16, y = var_10444_to_fp16)[name = string("aw_chunk_1009_cast_fp16")]; fp16 var_10446_to_fp16 = const()[name = string("op_10446_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1011_cast_fp16, y = var_10446_to_fp16)[name = string("aw_chunk_1011_cast_fp16")]; fp16 var_10448_to_fp16 = const()[name = string("op_10448_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1013_cast_fp16, y = var_10448_to_fp16)[name = string("aw_chunk_1013_cast_fp16")]; fp16 var_10450_to_fp16 = const()[name = string("op_10450_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1015_cast_fp16, y = var_10450_to_fp16)[name = string("aw_chunk_1015_cast_fp16")]; fp16 var_10452_to_fp16 = const()[name = string("op_10452_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1017_cast_fp16, y = var_10452_to_fp16)[name = string("aw_chunk_1017_cast_fp16")]; fp16 var_10454_to_fp16 = const()[name = string("op_10454_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1019_cast_fp16, y = var_10454_to_fp16)[name = string("aw_chunk_1019_cast_fp16")]; fp16 var_10456_to_fp16 = const()[name = string("op_10456_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1021_cast_fp16, y = var_10456_to_fp16)[name = string("aw_chunk_1021_cast_fp16")]; fp16 var_10458_to_fp16 = const()[name = string("op_10458_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1023_cast_fp16, y = var_10458_to_fp16)[name = string("aw_chunk_1023_cast_fp16")]; fp16 var_10460_to_fp16 = const()[name = string("op_10460_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1025_cast_fp16, y = var_10460_to_fp16)[name = string("aw_chunk_1025_cast_fp16")]; fp16 var_10462_to_fp16 = const()[name = string("op_10462_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1027_cast_fp16, y = var_10462_to_fp16)[name = string("aw_chunk_1027_cast_fp16")]; fp16 var_10464_to_fp16 = const()[name = string("op_10464_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1029_cast_fp16, y = var_10464_to_fp16)[name = string("aw_chunk_1029_cast_fp16")]; fp16 var_10466_to_fp16 = const()[name = string("op_10466_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1031_cast_fp16, y = var_10466_to_fp16)[name = string("aw_chunk_1031_cast_fp16")]; fp16 var_10468_to_fp16 = const()[name = string("op_10468_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1033_cast_fp16, y = var_10468_to_fp16)[name = string("aw_chunk_1033_cast_fp16")]; fp16 var_10470_to_fp16 = const()[name = string("op_10470_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1035_cast_fp16, y = var_10470_to_fp16)[name = string("aw_chunk_1035_cast_fp16")]; fp16 var_10472_to_fp16 = const()[name = string("op_10472_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1037_cast_fp16, y = var_10472_to_fp16)[name = string("aw_chunk_1037_cast_fp16")]; fp16 var_10474_to_fp16 = const()[name = string("op_10474_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1039_cast_fp16, y = var_10474_to_fp16)[name = string("aw_chunk_1039_cast_fp16")]; fp16 var_10476_to_fp16 = const()[name = string("op_10476_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1041_cast_fp16, y = var_10476_to_fp16)[name = string("aw_chunk_1041_cast_fp16")]; fp16 var_10478_to_fp16 = const()[name = string("op_10478_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1043_cast_fp16, y = var_10478_to_fp16)[name = string("aw_chunk_1043_cast_fp16")]; fp16 var_10480_to_fp16 = const()[name = string("op_10480_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1045_cast_fp16, y = var_10480_to_fp16)[name = string("aw_chunk_1045_cast_fp16")]; fp16 var_10482_to_fp16 = const()[name = string("op_10482_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1047_cast_fp16, y = var_10482_to_fp16)[name = string("aw_chunk_1047_cast_fp16")]; fp16 var_10484_to_fp16 = const()[name = string("op_10484_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1049_cast_fp16, y = var_10484_to_fp16)[name = string("aw_chunk_1049_cast_fp16")]; fp16 var_10486_to_fp16 = const()[name = string("op_10486_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1051_cast_fp16, y = var_10486_to_fp16)[name = string("aw_chunk_1051_cast_fp16")]; fp16 var_10488_to_fp16 = const()[name = string("op_10488_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1053_cast_fp16, y = var_10488_to_fp16)[name = string("aw_chunk_1053_cast_fp16")]; fp16 var_10490_to_fp16 = const()[name = string("op_10490_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1055_cast_fp16, y = var_10490_to_fp16)[name = string("aw_chunk_1055_cast_fp16")]; tensor var_10492_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_961_cast_fp16)[name = string("op_10492_cast_fp16")]; tensor var_10493_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_963_cast_fp16)[name = string("op_10493_cast_fp16")]; tensor var_10494_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_965_cast_fp16)[name = string("op_10494_cast_fp16")]; tensor var_10495_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_967_cast_fp16)[name = string("op_10495_cast_fp16")]; tensor var_10496_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_969_cast_fp16)[name = string("op_10496_cast_fp16")]; tensor var_10497_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_971_cast_fp16)[name = string("op_10497_cast_fp16")]; tensor var_10498_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_973_cast_fp16)[name = string("op_10498_cast_fp16")]; tensor var_10499_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_975_cast_fp16)[name = string("op_10499_cast_fp16")]; tensor var_10500_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_977_cast_fp16)[name = string("op_10500_cast_fp16")]; tensor var_10501_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_979_cast_fp16)[name = string("op_10501_cast_fp16")]; tensor var_10502_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_981_cast_fp16)[name = string("op_10502_cast_fp16")]; tensor var_10503_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_983_cast_fp16)[name = string("op_10503_cast_fp16")]; tensor var_10504_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_985_cast_fp16)[name = string("op_10504_cast_fp16")]; tensor var_10505_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_987_cast_fp16)[name = string("op_10505_cast_fp16")]; tensor var_10506_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_989_cast_fp16)[name = string("op_10506_cast_fp16")]; tensor var_10507_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_991_cast_fp16)[name = string("op_10507_cast_fp16")]; tensor var_10508_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_993_cast_fp16)[name = string("op_10508_cast_fp16")]; tensor var_10509_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_995_cast_fp16)[name = string("op_10509_cast_fp16")]; tensor var_10510_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_997_cast_fp16)[name = string("op_10510_cast_fp16")]; tensor var_10511_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_999_cast_fp16)[name = string("op_10511_cast_fp16")]; tensor var_10512_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1001_cast_fp16)[name = string("op_10512_cast_fp16")]; tensor var_10513_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1003_cast_fp16)[name = string("op_10513_cast_fp16")]; tensor var_10514_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1005_cast_fp16)[name = string("op_10514_cast_fp16")]; tensor var_10515_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1007_cast_fp16)[name = string("op_10515_cast_fp16")]; tensor var_10516_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1009_cast_fp16)[name = string("op_10516_cast_fp16")]; tensor var_10517_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1011_cast_fp16)[name = string("op_10517_cast_fp16")]; tensor var_10518_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1013_cast_fp16)[name = string("op_10518_cast_fp16")]; tensor var_10519_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1015_cast_fp16)[name = string("op_10519_cast_fp16")]; tensor var_10520_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1017_cast_fp16)[name = string("op_10520_cast_fp16")]; tensor var_10521_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1019_cast_fp16)[name = string("op_10521_cast_fp16")]; tensor var_10522_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1021_cast_fp16)[name = string("op_10522_cast_fp16")]; tensor var_10523_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1023_cast_fp16)[name = string("op_10523_cast_fp16")]; tensor var_10524_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1025_cast_fp16)[name = string("op_10524_cast_fp16")]; tensor var_10525_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1027_cast_fp16)[name = string("op_10525_cast_fp16")]; tensor var_10526_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1029_cast_fp16)[name = string("op_10526_cast_fp16")]; tensor var_10527_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1031_cast_fp16)[name = string("op_10527_cast_fp16")]; tensor var_10528_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1033_cast_fp16)[name = string("op_10528_cast_fp16")]; tensor var_10529_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1035_cast_fp16)[name = string("op_10529_cast_fp16")]; tensor var_10530_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1037_cast_fp16)[name = string("op_10530_cast_fp16")]; tensor var_10531_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1039_cast_fp16)[name = string("op_10531_cast_fp16")]; tensor var_10532_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1041_cast_fp16)[name = string("op_10532_cast_fp16")]; tensor var_10533_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1043_cast_fp16)[name = string("op_10533_cast_fp16")]; tensor var_10534_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1045_cast_fp16)[name = string("op_10534_cast_fp16")]; tensor var_10535_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1047_cast_fp16)[name = string("op_10535_cast_fp16")]; tensor var_10536_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1049_cast_fp16)[name = string("op_10536_cast_fp16")]; tensor var_10537_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1051_cast_fp16)[name = string("op_10537_cast_fp16")]; tensor var_10538_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1053_cast_fp16)[name = string("op_10538_cast_fp16")]; tensor var_10539_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1055_cast_fp16)[name = string("op_10539_cast_fp16")]; string var_10541_equation_0 = const()[name = string("op_10541_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10541_cast_fp16 = einsum(equation = var_10541_equation_0, values = (var_10253_cast_fp16, var_10492_cast_fp16))[name = string("op_10541_cast_fp16")]; string var_10543_equation_0 = const()[name = string("op_10543_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10543_cast_fp16 = einsum(equation = var_10543_equation_0, values = (var_10253_cast_fp16, var_10493_cast_fp16))[name = string("op_10543_cast_fp16")]; string var_10545_equation_0 = const()[name = string("op_10545_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10545_cast_fp16 = einsum(equation = var_10545_equation_0, values = (var_10253_cast_fp16, var_10494_cast_fp16))[name = string("op_10545_cast_fp16")]; string var_10547_equation_0 = const()[name = string("op_10547_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10547_cast_fp16 = einsum(equation = var_10547_equation_0, values = (var_10253_cast_fp16, var_10495_cast_fp16))[name = string("op_10547_cast_fp16")]; string var_10549_equation_0 = const()[name = string("op_10549_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10549_cast_fp16 = einsum(equation = var_10549_equation_0, values = (var_10257_cast_fp16, var_10496_cast_fp16))[name = string("op_10549_cast_fp16")]; string var_10551_equation_0 = const()[name = string("op_10551_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10551_cast_fp16 = einsum(equation = var_10551_equation_0, values = (var_10257_cast_fp16, var_10497_cast_fp16))[name = string("op_10551_cast_fp16")]; string var_10553_equation_0 = const()[name = string("op_10553_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10553_cast_fp16 = einsum(equation = var_10553_equation_0, values = (var_10257_cast_fp16, var_10498_cast_fp16))[name = string("op_10553_cast_fp16")]; string var_10555_equation_0 = const()[name = string("op_10555_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10555_cast_fp16 = einsum(equation = var_10555_equation_0, values = (var_10257_cast_fp16, var_10499_cast_fp16))[name = string("op_10555_cast_fp16")]; string var_10557_equation_0 = const()[name = string("op_10557_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10557_cast_fp16 = einsum(equation = var_10557_equation_0, values = (var_10261_cast_fp16, var_10500_cast_fp16))[name = string("op_10557_cast_fp16")]; string var_10559_equation_0 = const()[name = string("op_10559_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10559_cast_fp16 = einsum(equation = var_10559_equation_0, values = (var_10261_cast_fp16, var_10501_cast_fp16))[name = string("op_10559_cast_fp16")]; string var_10561_equation_0 = const()[name = string("op_10561_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10561_cast_fp16 = einsum(equation = var_10561_equation_0, values = (var_10261_cast_fp16, var_10502_cast_fp16))[name = string("op_10561_cast_fp16")]; string var_10563_equation_0 = const()[name = string("op_10563_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10563_cast_fp16 = einsum(equation = var_10563_equation_0, values = (var_10261_cast_fp16, var_10503_cast_fp16))[name = string("op_10563_cast_fp16")]; string var_10565_equation_0 = const()[name = string("op_10565_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10565_cast_fp16 = einsum(equation = var_10565_equation_0, values = (var_10265_cast_fp16, var_10504_cast_fp16))[name = string("op_10565_cast_fp16")]; string var_10567_equation_0 = const()[name = string("op_10567_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10567_cast_fp16 = einsum(equation = var_10567_equation_0, values = (var_10265_cast_fp16, var_10505_cast_fp16))[name = string("op_10567_cast_fp16")]; string var_10569_equation_0 = const()[name = string("op_10569_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10569_cast_fp16 = einsum(equation = var_10569_equation_0, values = (var_10265_cast_fp16, var_10506_cast_fp16))[name = string("op_10569_cast_fp16")]; string var_10571_equation_0 = const()[name = string("op_10571_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10571_cast_fp16 = einsum(equation = var_10571_equation_0, values = (var_10265_cast_fp16, var_10507_cast_fp16))[name = string("op_10571_cast_fp16")]; string var_10573_equation_0 = const()[name = string("op_10573_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10573_cast_fp16 = einsum(equation = var_10573_equation_0, values = (var_10269_cast_fp16, var_10508_cast_fp16))[name = string("op_10573_cast_fp16")]; string var_10575_equation_0 = const()[name = string("op_10575_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10575_cast_fp16 = einsum(equation = var_10575_equation_0, values = (var_10269_cast_fp16, var_10509_cast_fp16))[name = string("op_10575_cast_fp16")]; string var_10577_equation_0 = const()[name = string("op_10577_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10577_cast_fp16 = einsum(equation = var_10577_equation_0, values = (var_10269_cast_fp16, var_10510_cast_fp16))[name = string("op_10577_cast_fp16")]; string var_10579_equation_0 = const()[name = string("op_10579_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10579_cast_fp16 = einsum(equation = var_10579_equation_0, values = (var_10269_cast_fp16, var_10511_cast_fp16))[name = string("op_10579_cast_fp16")]; string var_10581_equation_0 = const()[name = string("op_10581_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10581_cast_fp16 = einsum(equation = var_10581_equation_0, values = (var_10273_cast_fp16, var_10512_cast_fp16))[name = string("op_10581_cast_fp16")]; string var_10583_equation_0 = const()[name = string("op_10583_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10583_cast_fp16 = einsum(equation = var_10583_equation_0, values = (var_10273_cast_fp16, var_10513_cast_fp16))[name = string("op_10583_cast_fp16")]; string var_10585_equation_0 = const()[name = string("op_10585_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10585_cast_fp16 = einsum(equation = var_10585_equation_0, values = (var_10273_cast_fp16, var_10514_cast_fp16))[name = string("op_10585_cast_fp16")]; string var_10587_equation_0 = const()[name = string("op_10587_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10587_cast_fp16 = einsum(equation = var_10587_equation_0, values = (var_10273_cast_fp16, var_10515_cast_fp16))[name = string("op_10587_cast_fp16")]; string var_10589_equation_0 = const()[name = string("op_10589_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10589_cast_fp16 = einsum(equation = var_10589_equation_0, values = (var_10277_cast_fp16, var_10516_cast_fp16))[name = string("op_10589_cast_fp16")]; string var_10591_equation_0 = const()[name = string("op_10591_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10591_cast_fp16 = einsum(equation = var_10591_equation_0, values = (var_10277_cast_fp16, var_10517_cast_fp16))[name = string("op_10591_cast_fp16")]; string var_10593_equation_0 = const()[name = string("op_10593_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10593_cast_fp16 = einsum(equation = var_10593_equation_0, values = (var_10277_cast_fp16, var_10518_cast_fp16))[name = string("op_10593_cast_fp16")]; string var_10595_equation_0 = const()[name = string("op_10595_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10595_cast_fp16 = einsum(equation = var_10595_equation_0, values = (var_10277_cast_fp16, var_10519_cast_fp16))[name = string("op_10595_cast_fp16")]; string var_10597_equation_0 = const()[name = string("op_10597_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10597_cast_fp16 = einsum(equation = var_10597_equation_0, values = (var_10281_cast_fp16, var_10520_cast_fp16))[name = string("op_10597_cast_fp16")]; string var_10599_equation_0 = const()[name = string("op_10599_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10599_cast_fp16 = einsum(equation = var_10599_equation_0, values = (var_10281_cast_fp16, var_10521_cast_fp16))[name = string("op_10599_cast_fp16")]; string var_10601_equation_0 = const()[name = string("op_10601_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10601_cast_fp16 = einsum(equation = var_10601_equation_0, values = (var_10281_cast_fp16, var_10522_cast_fp16))[name = string("op_10601_cast_fp16")]; string var_10603_equation_0 = const()[name = string("op_10603_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10603_cast_fp16 = einsum(equation = var_10603_equation_0, values = (var_10281_cast_fp16, var_10523_cast_fp16))[name = string("op_10603_cast_fp16")]; string var_10605_equation_0 = const()[name = string("op_10605_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10605_cast_fp16 = einsum(equation = var_10605_equation_0, values = (var_10285_cast_fp16, var_10524_cast_fp16))[name = string("op_10605_cast_fp16")]; string var_10607_equation_0 = const()[name = string("op_10607_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10607_cast_fp16 = einsum(equation = var_10607_equation_0, values = (var_10285_cast_fp16, var_10525_cast_fp16))[name = string("op_10607_cast_fp16")]; string var_10609_equation_0 = const()[name = string("op_10609_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10609_cast_fp16 = einsum(equation = var_10609_equation_0, values = (var_10285_cast_fp16, var_10526_cast_fp16))[name = string("op_10609_cast_fp16")]; string var_10611_equation_0 = const()[name = string("op_10611_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10611_cast_fp16 = einsum(equation = var_10611_equation_0, values = (var_10285_cast_fp16, var_10527_cast_fp16))[name = string("op_10611_cast_fp16")]; string var_10613_equation_0 = const()[name = string("op_10613_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10613_cast_fp16 = einsum(equation = var_10613_equation_0, values = (var_10289_cast_fp16, var_10528_cast_fp16))[name = string("op_10613_cast_fp16")]; string var_10615_equation_0 = const()[name = string("op_10615_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10615_cast_fp16 = einsum(equation = var_10615_equation_0, values = (var_10289_cast_fp16, var_10529_cast_fp16))[name = string("op_10615_cast_fp16")]; string var_10617_equation_0 = const()[name = string("op_10617_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10617_cast_fp16 = einsum(equation = var_10617_equation_0, values = (var_10289_cast_fp16, var_10530_cast_fp16))[name = string("op_10617_cast_fp16")]; string var_10619_equation_0 = const()[name = string("op_10619_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10619_cast_fp16 = einsum(equation = var_10619_equation_0, values = (var_10289_cast_fp16, var_10531_cast_fp16))[name = string("op_10619_cast_fp16")]; string var_10621_equation_0 = const()[name = string("op_10621_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10621_cast_fp16 = einsum(equation = var_10621_equation_0, values = (var_10293_cast_fp16, var_10532_cast_fp16))[name = string("op_10621_cast_fp16")]; string var_10623_equation_0 = const()[name = string("op_10623_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10623_cast_fp16 = einsum(equation = var_10623_equation_0, values = (var_10293_cast_fp16, var_10533_cast_fp16))[name = string("op_10623_cast_fp16")]; string var_10625_equation_0 = const()[name = string("op_10625_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10625_cast_fp16 = einsum(equation = var_10625_equation_0, values = (var_10293_cast_fp16, var_10534_cast_fp16))[name = string("op_10625_cast_fp16")]; string var_10627_equation_0 = const()[name = string("op_10627_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10627_cast_fp16 = einsum(equation = var_10627_equation_0, values = (var_10293_cast_fp16, var_10535_cast_fp16))[name = string("op_10627_cast_fp16")]; string var_10629_equation_0 = const()[name = string("op_10629_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10629_cast_fp16 = einsum(equation = var_10629_equation_0, values = (var_10297_cast_fp16, var_10536_cast_fp16))[name = string("op_10629_cast_fp16")]; string var_10631_equation_0 = const()[name = string("op_10631_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10631_cast_fp16 = einsum(equation = var_10631_equation_0, values = (var_10297_cast_fp16, var_10537_cast_fp16))[name = string("op_10631_cast_fp16")]; string var_10633_equation_0 = const()[name = string("op_10633_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10633_cast_fp16 = einsum(equation = var_10633_equation_0, values = (var_10297_cast_fp16, var_10538_cast_fp16))[name = string("op_10633_cast_fp16")]; string var_10635_equation_0 = const()[name = string("op_10635_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_10635_cast_fp16 = einsum(equation = var_10635_equation_0, values = (var_10297_cast_fp16, var_10539_cast_fp16))[name = string("op_10635_cast_fp16")]; bool var_10637_interleave_0 = const()[name = string("op_10637_interleave_0"), val = bool(false)]; tensor var_10637_cast_fp16 = concat(axis = var_9748, interleave = var_10637_interleave_0, values = (var_10541_cast_fp16, var_10543_cast_fp16, var_10545_cast_fp16, var_10547_cast_fp16))[name = string("op_10637_cast_fp16")]; bool var_10639_interleave_0 = const()[name = string("op_10639_interleave_0"), val = bool(false)]; tensor var_10639_cast_fp16 = concat(axis = var_9748, interleave = var_10639_interleave_0, values = (var_10549_cast_fp16, var_10551_cast_fp16, var_10553_cast_fp16, var_10555_cast_fp16))[name = string("op_10639_cast_fp16")]; bool var_10641_interleave_0 = const()[name = string("op_10641_interleave_0"), val = bool(false)]; tensor var_10641_cast_fp16 = concat(axis = var_9748, interleave = var_10641_interleave_0, values = (var_10557_cast_fp16, var_10559_cast_fp16, var_10561_cast_fp16, var_10563_cast_fp16))[name = string("op_10641_cast_fp16")]; bool var_10643_interleave_0 = const()[name = string("op_10643_interleave_0"), val = bool(false)]; tensor var_10643_cast_fp16 = concat(axis = var_9748, interleave = var_10643_interleave_0, values = (var_10565_cast_fp16, var_10567_cast_fp16, var_10569_cast_fp16, var_10571_cast_fp16))[name = string("op_10643_cast_fp16")]; bool var_10645_interleave_0 = const()[name = string("op_10645_interleave_0"), val = bool(false)]; tensor var_10645_cast_fp16 = concat(axis = var_9748, interleave = var_10645_interleave_0, values = (var_10573_cast_fp16, var_10575_cast_fp16, var_10577_cast_fp16, var_10579_cast_fp16))[name = string("op_10645_cast_fp16")]; bool var_10647_interleave_0 = const()[name = string("op_10647_interleave_0"), val = bool(false)]; tensor var_10647_cast_fp16 = concat(axis = var_9748, interleave = var_10647_interleave_0, values = (var_10581_cast_fp16, var_10583_cast_fp16, var_10585_cast_fp16, var_10587_cast_fp16))[name = string("op_10647_cast_fp16")]; bool var_10649_interleave_0 = const()[name = string("op_10649_interleave_0"), val = bool(false)]; tensor var_10649_cast_fp16 = concat(axis = var_9748, interleave = var_10649_interleave_0, values = (var_10589_cast_fp16, var_10591_cast_fp16, var_10593_cast_fp16, var_10595_cast_fp16))[name = string("op_10649_cast_fp16")]; bool var_10651_interleave_0 = const()[name = string("op_10651_interleave_0"), val = bool(false)]; tensor var_10651_cast_fp16 = concat(axis = var_9748, interleave = var_10651_interleave_0, values = (var_10597_cast_fp16, var_10599_cast_fp16, var_10601_cast_fp16, var_10603_cast_fp16))[name = string("op_10651_cast_fp16")]; bool var_10653_interleave_0 = const()[name = string("op_10653_interleave_0"), val = bool(false)]; tensor var_10653_cast_fp16 = concat(axis = var_9748, interleave = var_10653_interleave_0, values = (var_10605_cast_fp16, var_10607_cast_fp16, var_10609_cast_fp16, var_10611_cast_fp16))[name = string("op_10653_cast_fp16")]; bool var_10655_interleave_0 = const()[name = string("op_10655_interleave_0"), val = bool(false)]; tensor var_10655_cast_fp16 = concat(axis = var_9748, interleave = var_10655_interleave_0, values = (var_10613_cast_fp16, var_10615_cast_fp16, var_10617_cast_fp16, var_10619_cast_fp16))[name = string("op_10655_cast_fp16")]; bool var_10657_interleave_0 = const()[name = string("op_10657_interleave_0"), val = bool(false)]; tensor var_10657_cast_fp16 = concat(axis = var_9748, interleave = var_10657_interleave_0, values = (var_10621_cast_fp16, var_10623_cast_fp16, var_10625_cast_fp16, var_10627_cast_fp16))[name = string("op_10657_cast_fp16")]; bool var_10659_interleave_0 = const()[name = string("op_10659_interleave_0"), val = bool(false)]; tensor var_10659_cast_fp16 = concat(axis = var_9748, interleave = var_10659_interleave_0, values = (var_10629_cast_fp16, var_10631_cast_fp16, var_10633_cast_fp16, var_10635_cast_fp16))[name = string("op_10659_cast_fp16")]; bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)]; tensor input_81_cast_fp16 = concat(axis = var_9765, interleave = input_81_interleave_0, values = (var_10637_cast_fp16, var_10639_cast_fp16, var_10641_cast_fp16, var_10643_cast_fp16, var_10645_cast_fp16, var_10647_cast_fp16, var_10649_cast_fp16, var_10651_cast_fp16, var_10653_cast_fp16, var_10655_cast_fp16, var_10657_cast_fp16, var_10659_cast_fp16))[name = string("input_81_cast_fp16")]; string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151515456)))]; tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152695168)))]; tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; fp16 var_10678_to_fp16 = const()[name = string("op_10678_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_10678_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152696768)))]; tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152698368)))]; fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152699968)))]; tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157418624)))]; tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157424832)))]; tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162143488)))]; tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; int32 var_10707 = const()[name = string("op_10707"), val = int32(3)]; int32 var_10724 = const()[name = string("op_10724"), val = int32(1)]; tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; fp16 var_10741_to_fp16 = const()[name = string("op_10741_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_10741_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162145088)))]; tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162146688)))]; fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162148288)))]; tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163328000)))]; tensor query_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_cast_fp16")]; string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163329600)))]; tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_cast_fp16")]; string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164509312)))]; tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165689024)))]; tensor value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_cast_fp16")]; tensor var_10779_begin_0 = const()[name = string("op_10779_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10779_end_0 = const()[name = string("op_10779_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10779_end_mask_0 = const()[name = string("op_10779_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10779_cast_fp16 = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = query_cast_fp16)[name = string("op_10779_cast_fp16")]; tensor var_10783_begin_0 = const()[name = string("op_10783_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10783_end_0 = const()[name = string("op_10783_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_10783_end_mask_0 = const()[name = string("op_10783_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10783_cast_fp16 = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = query_cast_fp16)[name = string("op_10783_cast_fp16")]; tensor var_10787_begin_0 = const()[name = string("op_10787_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10787_end_0 = const()[name = string("op_10787_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_10787_end_mask_0 = const()[name = string("op_10787_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10787_cast_fp16 = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = query_cast_fp16)[name = string("op_10787_cast_fp16")]; tensor var_10791_begin_0 = const()[name = string("op_10791_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10791_end_0 = const()[name = string("op_10791_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_10791_end_mask_0 = const()[name = string("op_10791_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10791_cast_fp16 = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = query_cast_fp16)[name = string("op_10791_cast_fp16")]; tensor var_10795_begin_0 = const()[name = string("op_10795_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10795_end_0 = const()[name = string("op_10795_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_10795_end_mask_0 = const()[name = string("op_10795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10795_cast_fp16 = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = query_cast_fp16)[name = string("op_10795_cast_fp16")]; tensor var_10799_begin_0 = const()[name = string("op_10799_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10799_end_0 = const()[name = string("op_10799_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_10799_end_mask_0 = const()[name = string("op_10799_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10799_cast_fp16 = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = query_cast_fp16)[name = string("op_10799_cast_fp16")]; tensor var_10803_begin_0 = const()[name = string("op_10803_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10803_end_0 = const()[name = string("op_10803_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_10803_end_mask_0 = const()[name = string("op_10803_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10803_cast_fp16 = slice_by_index(begin = var_10803_begin_0, end = var_10803_end_0, end_mask = var_10803_end_mask_0, x = query_cast_fp16)[name = string("op_10803_cast_fp16")]; tensor var_10807_begin_0 = const()[name = string("op_10807_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10807_end_0 = const()[name = string("op_10807_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_10807_end_mask_0 = const()[name = string("op_10807_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10807_cast_fp16 = slice_by_index(begin = var_10807_begin_0, end = var_10807_end_0, end_mask = var_10807_end_mask_0, x = query_cast_fp16)[name = string("op_10807_cast_fp16")]; tensor var_10811_begin_0 = const()[name = string("op_10811_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10811_end_0 = const()[name = string("op_10811_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_10811_end_mask_0 = const()[name = string("op_10811_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10811_cast_fp16 = slice_by_index(begin = var_10811_begin_0, end = var_10811_end_0, end_mask = var_10811_end_mask_0, x = query_cast_fp16)[name = string("op_10811_cast_fp16")]; tensor var_10815_begin_0 = const()[name = string("op_10815_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10815_end_0 = const()[name = string("op_10815_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_10815_end_mask_0 = const()[name = string("op_10815_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10815_cast_fp16 = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = query_cast_fp16)[name = string("op_10815_cast_fp16")]; tensor var_10819_begin_0 = const()[name = string("op_10819_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_10819_end_0 = const()[name = string("op_10819_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_10819_end_mask_0 = const()[name = string("op_10819_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10819_cast_fp16 = slice_by_index(begin = var_10819_begin_0, end = var_10819_end_0, end_mask = var_10819_end_mask_0, x = query_cast_fp16)[name = string("op_10819_cast_fp16")]; tensor var_10823_begin_0 = const()[name = string("op_10823_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_10823_end_0 = const()[name = string("op_10823_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_10823_end_mask_0 = const()[name = string("op_10823_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10823_cast_fp16 = slice_by_index(begin = var_10823_begin_0, end = var_10823_end_0, end_mask = var_10823_end_mask_0, x = query_cast_fp16)[name = string("op_10823_cast_fp16")]; tensor var_10832_begin_0 = const()[name = string("op_10832_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10832_end_0 = const()[name = string("op_10832_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10832_end_mask_0 = const()[name = string("op_10832_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10832_cast_fp16 = slice_by_index(begin = var_10832_begin_0, end = var_10832_end_0, end_mask = var_10832_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10832_cast_fp16")]; tensor var_10839_begin_0 = const()[name = string("op_10839_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10839_end_0 = const()[name = string("op_10839_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10839_end_mask_0 = const()[name = string("op_10839_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10839_cast_fp16 = slice_by_index(begin = var_10839_begin_0, end = var_10839_end_0, end_mask = var_10839_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10839_cast_fp16")]; tensor var_10846_begin_0 = const()[name = string("op_10846_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10846_end_0 = const()[name = string("op_10846_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10846_end_mask_0 = const()[name = string("op_10846_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10846_cast_fp16 = slice_by_index(begin = var_10846_begin_0, end = var_10846_end_0, end_mask = var_10846_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10846_cast_fp16")]; tensor var_10853_begin_0 = const()[name = string("op_10853_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10853_end_0 = const()[name = string("op_10853_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10853_end_mask_0 = const()[name = string("op_10853_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10853_cast_fp16 = slice_by_index(begin = var_10853_begin_0, end = var_10853_end_0, end_mask = var_10853_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10853_cast_fp16")]; tensor var_10860_begin_0 = const()[name = string("op_10860_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10860_end_0 = const()[name = string("op_10860_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10860_end_mask_0 = const()[name = string("op_10860_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10860_cast_fp16 = slice_by_index(begin = var_10860_begin_0, end = var_10860_end_0, end_mask = var_10860_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10860_cast_fp16")]; tensor var_10867_begin_0 = const()[name = string("op_10867_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10867_end_0 = const()[name = string("op_10867_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10867_end_mask_0 = const()[name = string("op_10867_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10867_cast_fp16 = slice_by_index(begin = var_10867_begin_0, end = var_10867_end_0, end_mask = var_10867_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10867_cast_fp16")]; tensor var_10874_begin_0 = const()[name = string("op_10874_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10874_end_0 = const()[name = string("op_10874_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10874_end_mask_0 = const()[name = string("op_10874_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10874_cast_fp16 = slice_by_index(begin = var_10874_begin_0, end = var_10874_end_0, end_mask = var_10874_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10874_cast_fp16")]; tensor var_10881_begin_0 = const()[name = string("op_10881_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10881_end_0 = const()[name = string("op_10881_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10881_end_mask_0 = const()[name = string("op_10881_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10881_cast_fp16 = slice_by_index(begin = var_10881_begin_0, end = var_10881_end_0, end_mask = var_10881_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10881_cast_fp16")]; tensor var_10888_begin_0 = const()[name = string("op_10888_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10888_end_0 = const()[name = string("op_10888_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10888_end_mask_0 = const()[name = string("op_10888_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10888_cast_fp16 = slice_by_index(begin = var_10888_begin_0, end = var_10888_end_0, end_mask = var_10888_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10888_cast_fp16")]; tensor var_10895_begin_0 = const()[name = string("op_10895_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10895_end_0 = const()[name = string("op_10895_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10895_end_mask_0 = const()[name = string("op_10895_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10895_cast_fp16 = slice_by_index(begin = var_10895_begin_0, end = var_10895_end_0, end_mask = var_10895_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10895_cast_fp16")]; tensor var_10902_begin_0 = const()[name = string("op_10902_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10902_end_0 = const()[name = string("op_10902_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10902_end_mask_0 = const()[name = string("op_10902_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10902_cast_fp16 = slice_by_index(begin = var_10902_begin_0, end = var_10902_end_0, end_mask = var_10902_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10902_cast_fp16")]; tensor var_10909_begin_0 = const()[name = string("op_10909_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10909_end_0 = const()[name = string("op_10909_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10909_end_mask_0 = const()[name = string("op_10909_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10909_cast_fp16 = slice_by_index(begin = var_10909_begin_0, end = var_10909_end_0, end_mask = var_10909_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10909_cast_fp16")]; tensor var_10916_begin_0 = const()[name = string("op_10916_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10916_end_0 = const()[name = string("op_10916_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10916_end_mask_0 = const()[name = string("op_10916_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10916_cast_fp16 = slice_by_index(begin = var_10916_begin_0, end = var_10916_end_0, end_mask = var_10916_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10916_cast_fp16")]; tensor var_10923_begin_0 = const()[name = string("op_10923_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10923_end_0 = const()[name = string("op_10923_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10923_end_mask_0 = const()[name = string("op_10923_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10923_cast_fp16 = slice_by_index(begin = var_10923_begin_0, end = var_10923_end_0, end_mask = var_10923_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10923_cast_fp16")]; tensor var_10930_begin_0 = const()[name = string("op_10930_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10930_end_0 = const()[name = string("op_10930_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10930_end_mask_0 = const()[name = string("op_10930_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10930_cast_fp16 = slice_by_index(begin = var_10930_begin_0, end = var_10930_end_0, end_mask = var_10930_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10930_cast_fp16")]; tensor var_10937_begin_0 = const()[name = string("op_10937_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10937_end_0 = const()[name = string("op_10937_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10937_end_mask_0 = const()[name = string("op_10937_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10937_cast_fp16 = slice_by_index(begin = var_10937_begin_0, end = var_10937_end_0, end_mask = var_10937_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10937_cast_fp16")]; tensor var_10944_begin_0 = const()[name = string("op_10944_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10944_end_0 = const()[name = string("op_10944_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10944_end_mask_0 = const()[name = string("op_10944_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10944_cast_fp16 = slice_by_index(begin = var_10944_begin_0, end = var_10944_end_0, end_mask = var_10944_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10944_cast_fp16")]; tensor var_10951_begin_0 = const()[name = string("op_10951_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10951_end_0 = const()[name = string("op_10951_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10951_end_mask_0 = const()[name = string("op_10951_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10951_cast_fp16 = slice_by_index(begin = var_10951_begin_0, end = var_10951_end_0, end_mask = var_10951_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10951_cast_fp16")]; tensor var_10958_begin_0 = const()[name = string("op_10958_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10958_end_0 = const()[name = string("op_10958_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10958_end_mask_0 = const()[name = string("op_10958_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10958_cast_fp16 = slice_by_index(begin = var_10958_begin_0, end = var_10958_end_0, end_mask = var_10958_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10958_cast_fp16")]; tensor var_10965_begin_0 = const()[name = string("op_10965_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10965_end_0 = const()[name = string("op_10965_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10965_end_mask_0 = const()[name = string("op_10965_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10965_cast_fp16 = slice_by_index(begin = var_10965_begin_0, end = var_10965_end_0, end_mask = var_10965_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10965_cast_fp16")]; tensor var_10972_begin_0 = const()[name = string("op_10972_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10972_end_0 = const()[name = string("op_10972_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10972_end_mask_0 = const()[name = string("op_10972_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10972_cast_fp16 = slice_by_index(begin = var_10972_begin_0, end = var_10972_end_0, end_mask = var_10972_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10972_cast_fp16")]; tensor var_10979_begin_0 = const()[name = string("op_10979_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10979_end_0 = const()[name = string("op_10979_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10979_end_mask_0 = const()[name = string("op_10979_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10979_cast_fp16")]; tensor var_10986_begin_0 = const()[name = string("op_10986_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10986_end_0 = const()[name = string("op_10986_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10986_end_mask_0 = const()[name = string("op_10986_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10986_cast_fp16 = slice_by_index(begin = var_10986_begin_0, end = var_10986_end_0, end_mask = var_10986_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10986_cast_fp16")]; tensor var_10993_begin_0 = const()[name = string("op_10993_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10993_end_0 = const()[name = string("op_10993_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10993_end_mask_0 = const()[name = string("op_10993_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10993_cast_fp16 = slice_by_index(begin = var_10993_begin_0, end = var_10993_end_0, end_mask = var_10993_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10993_cast_fp16")]; tensor var_11000_begin_0 = const()[name = string("op_11000_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11000_end_0 = const()[name = string("op_11000_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11000_end_mask_0 = const()[name = string("op_11000_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11000_cast_fp16 = slice_by_index(begin = var_11000_begin_0, end = var_11000_end_0, end_mask = var_11000_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11000_cast_fp16")]; tensor var_11007_begin_0 = const()[name = string("op_11007_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11007_end_0 = const()[name = string("op_11007_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11007_end_mask_0 = const()[name = string("op_11007_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11007_cast_fp16")]; tensor var_11014_begin_0 = const()[name = string("op_11014_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11014_end_0 = const()[name = string("op_11014_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11014_end_mask_0 = const()[name = string("op_11014_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11014_cast_fp16 = slice_by_index(begin = var_11014_begin_0, end = var_11014_end_0, end_mask = var_11014_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11014_cast_fp16")]; tensor var_11021_begin_0 = const()[name = string("op_11021_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11021_end_0 = const()[name = string("op_11021_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11021_end_mask_0 = const()[name = string("op_11021_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11021_cast_fp16 = slice_by_index(begin = var_11021_begin_0, end = var_11021_end_0, end_mask = var_11021_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11021_cast_fp16")]; tensor var_11028_begin_0 = const()[name = string("op_11028_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11028_end_0 = const()[name = string("op_11028_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11028_end_mask_0 = const()[name = string("op_11028_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11028_cast_fp16 = slice_by_index(begin = var_11028_begin_0, end = var_11028_end_0, end_mask = var_11028_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11028_cast_fp16")]; tensor var_11035_begin_0 = const()[name = string("op_11035_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11035_end_0 = const()[name = string("op_11035_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11035_end_mask_0 = const()[name = string("op_11035_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11035_cast_fp16 = slice_by_index(begin = var_11035_begin_0, end = var_11035_end_0, end_mask = var_11035_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11035_cast_fp16")]; tensor var_11042_begin_0 = const()[name = string("op_11042_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11042_end_0 = const()[name = string("op_11042_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11042_end_mask_0 = const()[name = string("op_11042_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11042_cast_fp16 = slice_by_index(begin = var_11042_begin_0, end = var_11042_end_0, end_mask = var_11042_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11042_cast_fp16")]; tensor var_11049_begin_0 = const()[name = string("op_11049_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11049_end_0 = const()[name = string("op_11049_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11049_end_mask_0 = const()[name = string("op_11049_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11049_cast_fp16 = slice_by_index(begin = var_11049_begin_0, end = var_11049_end_0, end_mask = var_11049_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11049_cast_fp16")]; tensor var_11056_begin_0 = const()[name = string("op_11056_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11056_end_0 = const()[name = string("op_11056_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11056_end_mask_0 = const()[name = string("op_11056_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11056_cast_fp16 = slice_by_index(begin = var_11056_begin_0, end = var_11056_end_0, end_mask = var_11056_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11056_cast_fp16")]; tensor var_11063_begin_0 = const()[name = string("op_11063_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11063_end_0 = const()[name = string("op_11063_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11063_end_mask_0 = const()[name = string("op_11063_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11063_cast_fp16 = slice_by_index(begin = var_11063_begin_0, end = var_11063_end_0, end_mask = var_11063_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11063_cast_fp16")]; tensor var_11070_begin_0 = const()[name = string("op_11070_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11070_end_0 = const()[name = string("op_11070_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11070_end_mask_0 = const()[name = string("op_11070_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11070_cast_fp16 = slice_by_index(begin = var_11070_begin_0, end = var_11070_end_0, end_mask = var_11070_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11070_cast_fp16")]; tensor var_11077_begin_0 = const()[name = string("op_11077_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11077_end_0 = const()[name = string("op_11077_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11077_end_mask_0 = const()[name = string("op_11077_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11077_cast_fp16 = slice_by_index(begin = var_11077_begin_0, end = var_11077_end_0, end_mask = var_11077_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11077_cast_fp16")]; tensor var_11084_begin_0 = const()[name = string("op_11084_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11084_end_0 = const()[name = string("op_11084_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11084_end_mask_0 = const()[name = string("op_11084_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11084_cast_fp16 = slice_by_index(begin = var_11084_begin_0, end = var_11084_end_0, end_mask = var_11084_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11084_cast_fp16")]; tensor var_11091_begin_0 = const()[name = string("op_11091_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11091_end_0 = const()[name = string("op_11091_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11091_end_mask_0 = const()[name = string("op_11091_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11091_cast_fp16 = slice_by_index(begin = var_11091_begin_0, end = var_11091_end_0, end_mask = var_11091_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11091_cast_fp16")]; tensor var_11098_begin_0 = const()[name = string("op_11098_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11098_end_0 = const()[name = string("op_11098_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11098_end_mask_0 = const()[name = string("op_11098_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11098_cast_fp16 = slice_by_index(begin = var_11098_begin_0, end = var_11098_end_0, end_mask = var_11098_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11098_cast_fp16")]; tensor var_11105_begin_0 = const()[name = string("op_11105_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11105_end_0 = const()[name = string("op_11105_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11105_end_mask_0 = const()[name = string("op_11105_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11105_cast_fp16 = slice_by_index(begin = var_11105_begin_0, end = var_11105_end_0, end_mask = var_11105_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11105_cast_fp16")]; tensor var_11112_begin_0 = const()[name = string("op_11112_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11112_end_0 = const()[name = string("op_11112_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11112_end_mask_0 = const()[name = string("op_11112_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11112_cast_fp16 = slice_by_index(begin = var_11112_begin_0, end = var_11112_end_0, end_mask = var_11112_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11112_cast_fp16")]; tensor var_11119_begin_0 = const()[name = string("op_11119_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11119_end_0 = const()[name = string("op_11119_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11119_end_mask_0 = const()[name = string("op_11119_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11119_cast_fp16 = slice_by_index(begin = var_11119_begin_0, end = var_11119_end_0, end_mask = var_11119_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11119_cast_fp16")]; tensor var_11126_begin_0 = const()[name = string("op_11126_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11126_end_0 = const()[name = string("op_11126_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11126_end_mask_0 = const()[name = string("op_11126_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11126_cast_fp16 = slice_by_index(begin = var_11126_begin_0, end = var_11126_end_0, end_mask = var_11126_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11126_cast_fp16")]; tensor var_11133_begin_0 = const()[name = string("op_11133_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11133_end_0 = const()[name = string("op_11133_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11133_end_mask_0 = const()[name = string("op_11133_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11133_cast_fp16 = slice_by_index(begin = var_11133_begin_0, end = var_11133_end_0, end_mask = var_11133_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11133_cast_fp16")]; tensor var_11140_begin_0 = const()[name = string("op_11140_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11140_end_0 = const()[name = string("op_11140_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11140_end_mask_0 = const()[name = string("op_11140_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11140_cast_fp16 = slice_by_index(begin = var_11140_begin_0, end = var_11140_end_0, end_mask = var_11140_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11140_cast_fp16")]; tensor var_11147_begin_0 = const()[name = string("op_11147_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11147_end_0 = const()[name = string("op_11147_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11147_end_mask_0 = const()[name = string("op_11147_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11147_cast_fp16 = slice_by_index(begin = var_11147_begin_0, end = var_11147_end_0, end_mask = var_11147_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11147_cast_fp16")]; tensor var_11154_begin_0 = const()[name = string("op_11154_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11154_end_0 = const()[name = string("op_11154_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11154_end_mask_0 = const()[name = string("op_11154_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11154_cast_fp16 = slice_by_index(begin = var_11154_begin_0, end = var_11154_end_0, end_mask = var_11154_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11154_cast_fp16")]; tensor var_11161_begin_0 = const()[name = string("op_11161_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11161_end_0 = const()[name = string("op_11161_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11161_end_mask_0 = const()[name = string("op_11161_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11161_cast_fp16 = slice_by_index(begin = var_11161_begin_0, end = var_11161_end_0, end_mask = var_11161_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11161_cast_fp16")]; tensor k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11166_begin_0 = const()[name = string("op_11166_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11166_end_0 = const()[name = string("op_11166_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_11166_end_mask_0 = const()[name = string("op_11166_end_mask_0"), val = tensor([true, true, true, false])]; tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; tensor var_11166_cast_fp16 = slice_by_index(begin = var_11166_begin_0, end = var_11166_end_0, end_mask = var_11166_end_mask_0, x = k_23_cast_fp16)[name = string("op_11166_cast_fp16")]; tensor var_11170_begin_0 = const()[name = string("op_11170_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11170_end_0 = const()[name = string("op_11170_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_11170_end_mask_0 = const()[name = string("op_11170_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11170_cast_fp16 = slice_by_index(begin = var_11170_begin_0, end = var_11170_end_0, end_mask = var_11170_end_mask_0, x = k_23_cast_fp16)[name = string("op_11170_cast_fp16")]; tensor var_11174_begin_0 = const()[name = string("op_11174_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11174_end_0 = const()[name = string("op_11174_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_11174_end_mask_0 = const()[name = string("op_11174_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11174_cast_fp16 = slice_by_index(begin = var_11174_begin_0, end = var_11174_end_0, end_mask = var_11174_end_mask_0, x = k_23_cast_fp16)[name = string("op_11174_cast_fp16")]; tensor var_11178_begin_0 = const()[name = string("op_11178_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11178_end_0 = const()[name = string("op_11178_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_11178_end_mask_0 = const()[name = string("op_11178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11178_cast_fp16 = slice_by_index(begin = var_11178_begin_0, end = var_11178_end_0, end_mask = var_11178_end_mask_0, x = k_23_cast_fp16)[name = string("op_11178_cast_fp16")]; tensor var_11182_begin_0 = const()[name = string("op_11182_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11182_end_0 = const()[name = string("op_11182_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_11182_end_mask_0 = const()[name = string("op_11182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11182_cast_fp16 = slice_by_index(begin = var_11182_begin_0, end = var_11182_end_0, end_mask = var_11182_end_mask_0, x = k_23_cast_fp16)[name = string("op_11182_cast_fp16")]; tensor var_11186_begin_0 = const()[name = string("op_11186_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_11186_end_0 = const()[name = string("op_11186_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_11186_end_mask_0 = const()[name = string("op_11186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11186_cast_fp16 = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = k_23_cast_fp16)[name = string("op_11186_cast_fp16")]; tensor var_11190_begin_0 = const()[name = string("op_11190_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_11190_end_0 = const()[name = string("op_11190_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_11190_end_mask_0 = const()[name = string("op_11190_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11190_cast_fp16 = slice_by_index(begin = var_11190_begin_0, end = var_11190_end_0, end_mask = var_11190_end_mask_0, x = k_23_cast_fp16)[name = string("op_11190_cast_fp16")]; tensor var_11194_begin_0 = const()[name = string("op_11194_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_11194_end_0 = const()[name = string("op_11194_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_11194_end_mask_0 = const()[name = string("op_11194_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11194_cast_fp16 = slice_by_index(begin = var_11194_begin_0, end = var_11194_end_0, end_mask = var_11194_end_mask_0, x = k_23_cast_fp16)[name = string("op_11194_cast_fp16")]; tensor var_11198_begin_0 = const()[name = string("op_11198_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11198_end_0 = const()[name = string("op_11198_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_11198_end_mask_0 = const()[name = string("op_11198_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11198_cast_fp16 = slice_by_index(begin = var_11198_begin_0, end = var_11198_end_0, end_mask = var_11198_end_mask_0, x = k_23_cast_fp16)[name = string("op_11198_cast_fp16")]; tensor var_11202_begin_0 = const()[name = string("op_11202_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_11202_end_0 = const()[name = string("op_11202_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_11202_end_mask_0 = const()[name = string("op_11202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11202_cast_fp16 = slice_by_index(begin = var_11202_begin_0, end = var_11202_end_0, end_mask = var_11202_end_mask_0, x = k_23_cast_fp16)[name = string("op_11202_cast_fp16")]; tensor var_11206_begin_0 = const()[name = string("op_11206_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_11206_end_0 = const()[name = string("op_11206_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_11206_end_mask_0 = const()[name = string("op_11206_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11206_cast_fp16 = slice_by_index(begin = var_11206_begin_0, end = var_11206_end_0, end_mask = var_11206_end_mask_0, x = k_23_cast_fp16)[name = string("op_11206_cast_fp16")]; tensor var_11210_begin_0 = const()[name = string("op_11210_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_11210_end_0 = const()[name = string("op_11210_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_11210_end_mask_0 = const()[name = string("op_11210_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11210_cast_fp16 = slice_by_index(begin = var_11210_begin_0, end = var_11210_end_0, end_mask = var_11210_end_mask_0, x = k_23_cast_fp16)[name = string("op_11210_cast_fp16")]; tensor var_11212_begin_0 = const()[name = string("op_11212_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11212_end_0 = const()[name = string("op_11212_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11212_end_mask_0 = const()[name = string("op_11212_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11212_cast_fp16 = slice_by_index(begin = var_11212_begin_0, end = var_11212_end_0, end_mask = var_11212_end_mask_0, x = value_cast_fp16)[name = string("op_11212_cast_fp16")]; tensor var_11216_begin_0 = const()[name = string("op_11216_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11216_end_0 = const()[name = string("op_11216_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_11216_end_mask_0 = const()[name = string("op_11216_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11216_cast_fp16 = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = value_cast_fp16)[name = string("op_11216_cast_fp16")]; tensor var_11220_begin_0 = const()[name = string("op_11220_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11220_end_0 = const()[name = string("op_11220_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_11220_end_mask_0 = const()[name = string("op_11220_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11220_cast_fp16 = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = value_cast_fp16)[name = string("op_11220_cast_fp16")]; tensor var_11224_begin_0 = const()[name = string("op_11224_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11224_end_0 = const()[name = string("op_11224_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_11224_end_mask_0 = const()[name = string("op_11224_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11224_cast_fp16 = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = value_cast_fp16)[name = string("op_11224_cast_fp16")]; tensor var_11228_begin_0 = const()[name = string("op_11228_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11228_end_0 = const()[name = string("op_11228_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_11228_end_mask_0 = const()[name = string("op_11228_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = value_cast_fp16)[name = string("op_11228_cast_fp16")]; tensor var_11232_begin_0 = const()[name = string("op_11232_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_11232_end_0 = const()[name = string("op_11232_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_11232_end_mask_0 = const()[name = string("op_11232_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = value_cast_fp16)[name = string("op_11232_cast_fp16")]; tensor var_11236_begin_0 = const()[name = string("op_11236_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_11236_end_0 = const()[name = string("op_11236_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_11236_end_mask_0 = const()[name = string("op_11236_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11236_cast_fp16 = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = value_cast_fp16)[name = string("op_11236_cast_fp16")]; tensor var_11240_begin_0 = const()[name = string("op_11240_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_11240_end_0 = const()[name = string("op_11240_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_11240_end_mask_0 = const()[name = string("op_11240_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11240_cast_fp16 = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = value_cast_fp16)[name = string("op_11240_cast_fp16")]; tensor var_11244_begin_0 = const()[name = string("op_11244_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_11244_end_0 = const()[name = string("op_11244_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_11244_end_mask_0 = const()[name = string("op_11244_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11244_cast_fp16 = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = value_cast_fp16)[name = string("op_11244_cast_fp16")]; tensor var_11248_begin_0 = const()[name = string("op_11248_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_11248_end_0 = const()[name = string("op_11248_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_11248_end_mask_0 = const()[name = string("op_11248_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11248_cast_fp16 = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = value_cast_fp16)[name = string("op_11248_cast_fp16")]; tensor var_11252_begin_0 = const()[name = string("op_11252_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_11252_end_0 = const()[name = string("op_11252_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_11252_end_mask_0 = const()[name = string("op_11252_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11252_cast_fp16 = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = value_cast_fp16)[name = string("op_11252_cast_fp16")]; tensor var_11256_begin_0 = const()[name = string("op_11256_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_11256_end_0 = const()[name = string("op_11256_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_11256_end_mask_0 = const()[name = string("op_11256_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = value_cast_fp16)[name = string("op_11256_cast_fp16")]; string _SplitHeadsQ__mh_w_1057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1057_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1057_equation_0, values = (var_11166_cast_fp16, var_10832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1057_cast_fp16")]; string _SplitHeadsQ__mh_w_1059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1059_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1059_equation_0, values = (var_11166_cast_fp16, var_10839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1059_cast_fp16")]; string _SplitHeadsQ__mh_w_1061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1061_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1061_equation_0, values = (var_11166_cast_fp16, var_10846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1061_cast_fp16")]; string _SplitHeadsQ__mh_w_1063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1063_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1063_equation_0, values = (var_11166_cast_fp16, var_10853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1063_cast_fp16")]; string _SplitHeadsQ__mh_w_1065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1065_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1065_equation_0, values = (var_11170_cast_fp16, var_10860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1065_cast_fp16")]; string _SplitHeadsQ__mh_w_1067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1067_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1067_equation_0, values = (var_11170_cast_fp16, var_10867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1067_cast_fp16")]; string _SplitHeadsQ__mh_w_1069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1069_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1069_equation_0, values = (var_11170_cast_fp16, var_10874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1069_cast_fp16")]; string _SplitHeadsQ__mh_w_1071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1071_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1071_equation_0, values = (var_11170_cast_fp16, var_10881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1071_cast_fp16")]; string _SplitHeadsQ__mh_w_1073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1073_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1073_equation_0, values = (var_11174_cast_fp16, var_10888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1073_cast_fp16")]; string _SplitHeadsQ__mh_w_1075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1075_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1075_equation_0, values = (var_11174_cast_fp16, var_10895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1075_cast_fp16")]; string _SplitHeadsQ__mh_w_1077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1077_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1077_equation_0, values = (var_11174_cast_fp16, var_10902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1077_cast_fp16")]; string _SplitHeadsQ__mh_w_1079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1079_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1079_equation_0, values = (var_11174_cast_fp16, var_10909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1079_cast_fp16")]; string _SplitHeadsQ__mh_w_1081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1081_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1081_equation_0, values = (var_11178_cast_fp16, var_10916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1081_cast_fp16")]; string _SplitHeadsQ__mh_w_1083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1083_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1083_equation_0, values = (var_11178_cast_fp16, var_10923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1083_cast_fp16")]; string _SplitHeadsQ__mh_w_1085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1085_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1085_equation_0, values = (var_11178_cast_fp16, var_10930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1085_cast_fp16")]; string _SplitHeadsQ__mh_w_1087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1087_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1087_equation_0, values = (var_11178_cast_fp16, var_10937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1087_cast_fp16")]; string _SplitHeadsQ__mh_w_1089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1089_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1089_equation_0, values = (var_11182_cast_fp16, var_10944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1089_cast_fp16")]; string _SplitHeadsQ__mh_w_1091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1091_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1091_equation_0, values = (var_11182_cast_fp16, var_10951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1091_cast_fp16")]; string _SplitHeadsQ__mh_w_1093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1093_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1093_equation_0, values = (var_11182_cast_fp16, var_10958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1093_cast_fp16")]; string _SplitHeadsQ__mh_w_1095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1095_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1095_equation_0, values = (var_11182_cast_fp16, var_10965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1095_cast_fp16")]; string _SplitHeadsQ__mh_w_1097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1097_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1097_equation_0, values = (var_11186_cast_fp16, var_10972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1097_cast_fp16")]; string _SplitHeadsQ__mh_w_1099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1099_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1099_equation_0, values = (var_11186_cast_fp16, var_10979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1099_cast_fp16")]; string _SplitHeadsQ__mh_w_1101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1101_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1101_equation_0, values = (var_11186_cast_fp16, var_10986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1101_cast_fp16")]; string _SplitHeadsQ__mh_w_1103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1103_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1103_equation_0, values = (var_11186_cast_fp16, var_10993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1103_cast_fp16")]; string _SplitHeadsQ__mh_w_1105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1105_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1105_equation_0, values = (var_11190_cast_fp16, var_11000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1105_cast_fp16")]; string _SplitHeadsQ__mh_w_1107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1107_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1107_equation_0, values = (var_11190_cast_fp16, var_11007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1107_cast_fp16")]; string _SplitHeadsQ__mh_w_1109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1109_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1109_equation_0, values = (var_11190_cast_fp16, var_11014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1109_cast_fp16")]; string _SplitHeadsQ__mh_w_1111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1111_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1111_equation_0, values = (var_11190_cast_fp16, var_11021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1111_cast_fp16")]; string _SplitHeadsQ__mh_w_1113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1113_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1113_equation_0, values = (var_11194_cast_fp16, var_11028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1113_cast_fp16")]; string _SplitHeadsQ__mh_w_1115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1115_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1115_equation_0, values = (var_11194_cast_fp16, var_11035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1115_cast_fp16")]; string _SplitHeadsQ__mh_w_1117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1117_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1117_equation_0, values = (var_11194_cast_fp16, var_11042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1117_cast_fp16")]; string _SplitHeadsQ__mh_w_1119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1119_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1119_equation_0, values = (var_11194_cast_fp16, var_11049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1119_cast_fp16")]; string _SplitHeadsQ__mh_w_1121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1121_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1121_equation_0, values = (var_11198_cast_fp16, var_11056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1121_cast_fp16")]; string _SplitHeadsQ__mh_w_1123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1123_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1123_equation_0, values = (var_11198_cast_fp16, var_11063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1123_cast_fp16")]; string _SplitHeadsQ__mh_w_1125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1125_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1125_equation_0, values = (var_11198_cast_fp16, var_11070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1125_cast_fp16")]; string _SplitHeadsQ__mh_w_1127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1127_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1127_equation_0, values = (var_11198_cast_fp16, var_11077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1127_cast_fp16")]; string _SplitHeadsQ__mh_w_1129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1129_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1129_equation_0, values = (var_11202_cast_fp16, var_11084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1129_cast_fp16")]; string _SplitHeadsQ__mh_w_1131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1131_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1131_equation_0, values = (var_11202_cast_fp16, var_11091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1131_cast_fp16")]; string _SplitHeadsQ__mh_w_1133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1133_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1133_equation_0, values = (var_11202_cast_fp16, var_11098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1133_cast_fp16")]; string _SplitHeadsQ__mh_w_1135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1135_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1135_equation_0, values = (var_11202_cast_fp16, var_11105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1135_cast_fp16")]; string _SplitHeadsQ__mh_w_1137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1137_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1137_equation_0, values = (var_11206_cast_fp16, var_11112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1137_cast_fp16")]; string _SplitHeadsQ__mh_w_1139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1139_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1139_equation_0, values = (var_11206_cast_fp16, var_11119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1139_cast_fp16")]; string _SplitHeadsQ__mh_w_1141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1141_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1141_equation_0, values = (var_11206_cast_fp16, var_11126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1141_cast_fp16")]; string _SplitHeadsQ__mh_w_1143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1143_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1143_equation_0, values = (var_11206_cast_fp16, var_11133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1143_cast_fp16")]; string _SplitHeadsQ__mh_w_1145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1145_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1145_equation_0, values = (var_11210_cast_fp16, var_11140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1145_cast_fp16")]; string _SplitHeadsQ__mh_w_1147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1147_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1147_equation_0, values = (var_11210_cast_fp16, var_11147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1147_cast_fp16")]; string _SplitHeadsQ__mh_w_1149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1149_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_1149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1149_equation_0, values = (var_11210_cast_fp16, var_11154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1149_cast_fp16")]; string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_11210_cast_fp16, var_11161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; fp16 var_11355_to_fp16 = const()[name = string("op_11355_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1057_cast_fp16, y = var_11355_to_fp16)[name = string("aw_chunk_1057_cast_fp16")]; fp16 var_11357_to_fp16 = const()[name = string("op_11357_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1059_cast_fp16, y = var_11357_to_fp16)[name = string("aw_chunk_1059_cast_fp16")]; fp16 var_11359_to_fp16 = const()[name = string("op_11359_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1061_cast_fp16, y = var_11359_to_fp16)[name = string("aw_chunk_1061_cast_fp16")]; fp16 var_11361_to_fp16 = const()[name = string("op_11361_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1063_cast_fp16, y = var_11361_to_fp16)[name = string("aw_chunk_1063_cast_fp16")]; fp16 var_11363_to_fp16 = const()[name = string("op_11363_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1065_cast_fp16, y = var_11363_to_fp16)[name = string("aw_chunk_1065_cast_fp16")]; fp16 var_11365_to_fp16 = const()[name = string("op_11365_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1067_cast_fp16, y = var_11365_to_fp16)[name = string("aw_chunk_1067_cast_fp16")]; fp16 var_11367_to_fp16 = const()[name = string("op_11367_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1069_cast_fp16, y = var_11367_to_fp16)[name = string("aw_chunk_1069_cast_fp16")]; fp16 var_11369_to_fp16 = const()[name = string("op_11369_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1071_cast_fp16, y = var_11369_to_fp16)[name = string("aw_chunk_1071_cast_fp16")]; fp16 var_11371_to_fp16 = const()[name = string("op_11371_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1073_cast_fp16, y = var_11371_to_fp16)[name = string("aw_chunk_1073_cast_fp16")]; fp16 var_11373_to_fp16 = const()[name = string("op_11373_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1075_cast_fp16, y = var_11373_to_fp16)[name = string("aw_chunk_1075_cast_fp16")]; fp16 var_11375_to_fp16 = const()[name = string("op_11375_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1077_cast_fp16, y = var_11375_to_fp16)[name = string("aw_chunk_1077_cast_fp16")]; fp16 var_11377_to_fp16 = const()[name = string("op_11377_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1079_cast_fp16, y = var_11377_to_fp16)[name = string("aw_chunk_1079_cast_fp16")]; fp16 var_11379_to_fp16 = const()[name = string("op_11379_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1081_cast_fp16, y = var_11379_to_fp16)[name = string("aw_chunk_1081_cast_fp16")]; fp16 var_11381_to_fp16 = const()[name = string("op_11381_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1083_cast_fp16, y = var_11381_to_fp16)[name = string("aw_chunk_1083_cast_fp16")]; fp16 var_11383_to_fp16 = const()[name = string("op_11383_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1085_cast_fp16, y = var_11383_to_fp16)[name = string("aw_chunk_1085_cast_fp16")]; fp16 var_11385_to_fp16 = const()[name = string("op_11385_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1087_cast_fp16, y = var_11385_to_fp16)[name = string("aw_chunk_1087_cast_fp16")]; fp16 var_11387_to_fp16 = const()[name = string("op_11387_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1089_cast_fp16, y = var_11387_to_fp16)[name = string("aw_chunk_1089_cast_fp16")]; fp16 var_11389_to_fp16 = const()[name = string("op_11389_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1091_cast_fp16, y = var_11389_to_fp16)[name = string("aw_chunk_1091_cast_fp16")]; fp16 var_11391_to_fp16 = const()[name = string("op_11391_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1093_cast_fp16, y = var_11391_to_fp16)[name = string("aw_chunk_1093_cast_fp16")]; fp16 var_11393_to_fp16 = const()[name = string("op_11393_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1095_cast_fp16, y = var_11393_to_fp16)[name = string("aw_chunk_1095_cast_fp16")]; fp16 var_11395_to_fp16 = const()[name = string("op_11395_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1097_cast_fp16, y = var_11395_to_fp16)[name = string("aw_chunk_1097_cast_fp16")]; fp16 var_11397_to_fp16 = const()[name = string("op_11397_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1099_cast_fp16, y = var_11397_to_fp16)[name = string("aw_chunk_1099_cast_fp16")]; fp16 var_11399_to_fp16 = const()[name = string("op_11399_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1101_cast_fp16, y = var_11399_to_fp16)[name = string("aw_chunk_1101_cast_fp16")]; fp16 var_11401_to_fp16 = const()[name = string("op_11401_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1103_cast_fp16, y = var_11401_to_fp16)[name = string("aw_chunk_1103_cast_fp16")]; fp16 var_11403_to_fp16 = const()[name = string("op_11403_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1105_cast_fp16, y = var_11403_to_fp16)[name = string("aw_chunk_1105_cast_fp16")]; fp16 var_11405_to_fp16 = const()[name = string("op_11405_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1107_cast_fp16, y = var_11405_to_fp16)[name = string("aw_chunk_1107_cast_fp16")]; fp16 var_11407_to_fp16 = const()[name = string("op_11407_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1109_cast_fp16, y = var_11407_to_fp16)[name = string("aw_chunk_1109_cast_fp16")]; fp16 var_11409_to_fp16 = const()[name = string("op_11409_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1111_cast_fp16, y = var_11409_to_fp16)[name = string("aw_chunk_1111_cast_fp16")]; fp16 var_11411_to_fp16 = const()[name = string("op_11411_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1113_cast_fp16, y = var_11411_to_fp16)[name = string("aw_chunk_1113_cast_fp16")]; fp16 var_11413_to_fp16 = const()[name = string("op_11413_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1115_cast_fp16, y = var_11413_to_fp16)[name = string("aw_chunk_1115_cast_fp16")]; fp16 var_11415_to_fp16 = const()[name = string("op_11415_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1117_cast_fp16, y = var_11415_to_fp16)[name = string("aw_chunk_1117_cast_fp16")]; fp16 var_11417_to_fp16 = const()[name = string("op_11417_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1119_cast_fp16, y = var_11417_to_fp16)[name = string("aw_chunk_1119_cast_fp16")]; fp16 var_11419_to_fp16 = const()[name = string("op_11419_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1121_cast_fp16, y = var_11419_to_fp16)[name = string("aw_chunk_1121_cast_fp16")]; fp16 var_11421_to_fp16 = const()[name = string("op_11421_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1123_cast_fp16, y = var_11421_to_fp16)[name = string("aw_chunk_1123_cast_fp16")]; fp16 var_11423_to_fp16 = const()[name = string("op_11423_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1125_cast_fp16, y = var_11423_to_fp16)[name = string("aw_chunk_1125_cast_fp16")]; fp16 var_11425_to_fp16 = const()[name = string("op_11425_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1127_cast_fp16, y = var_11425_to_fp16)[name = string("aw_chunk_1127_cast_fp16")]; fp16 var_11427_to_fp16 = const()[name = string("op_11427_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1129_cast_fp16, y = var_11427_to_fp16)[name = string("aw_chunk_1129_cast_fp16")]; fp16 var_11429_to_fp16 = const()[name = string("op_11429_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1131_cast_fp16, y = var_11429_to_fp16)[name = string("aw_chunk_1131_cast_fp16")]; fp16 var_11431_to_fp16 = const()[name = string("op_11431_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1133_cast_fp16, y = var_11431_to_fp16)[name = string("aw_chunk_1133_cast_fp16")]; fp16 var_11433_to_fp16 = const()[name = string("op_11433_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1135_cast_fp16, y = var_11433_to_fp16)[name = string("aw_chunk_1135_cast_fp16")]; fp16 var_11435_to_fp16 = const()[name = string("op_11435_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1137_cast_fp16, y = var_11435_to_fp16)[name = string("aw_chunk_1137_cast_fp16")]; fp16 var_11437_to_fp16 = const()[name = string("op_11437_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1139_cast_fp16, y = var_11437_to_fp16)[name = string("aw_chunk_1139_cast_fp16")]; fp16 var_11439_to_fp16 = const()[name = string("op_11439_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1141_cast_fp16, y = var_11439_to_fp16)[name = string("aw_chunk_1141_cast_fp16")]; fp16 var_11441_to_fp16 = const()[name = string("op_11441_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1143_cast_fp16, y = var_11441_to_fp16)[name = string("aw_chunk_1143_cast_fp16")]; fp16 var_11443_to_fp16 = const()[name = string("op_11443_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1145_cast_fp16, y = var_11443_to_fp16)[name = string("aw_chunk_1145_cast_fp16")]; fp16 var_11445_to_fp16 = const()[name = string("op_11445_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1147_cast_fp16, y = var_11445_to_fp16)[name = string("aw_chunk_1147_cast_fp16")]; fp16 var_11447_to_fp16 = const()[name = string("op_11447_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_1149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1149_cast_fp16, y = var_11447_to_fp16)[name = string("aw_chunk_1149_cast_fp16")]; fp16 var_11449_to_fp16 = const()[name = string("op_11449_to_fp16"), val = fp16(0x1p-3)]; tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_11449_to_fp16)[name = string("aw_chunk_cast_fp16")]; tensor var_11451_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1057_cast_fp16)[name = string("op_11451_cast_fp16")]; tensor var_11452_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1059_cast_fp16)[name = string("op_11452_cast_fp16")]; tensor var_11453_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1061_cast_fp16)[name = string("op_11453_cast_fp16")]; tensor var_11454_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1063_cast_fp16)[name = string("op_11454_cast_fp16")]; tensor var_11455_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1065_cast_fp16)[name = string("op_11455_cast_fp16")]; tensor var_11456_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1067_cast_fp16)[name = string("op_11456_cast_fp16")]; tensor var_11457_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1069_cast_fp16)[name = string("op_11457_cast_fp16")]; tensor var_11458_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1071_cast_fp16)[name = string("op_11458_cast_fp16")]; tensor var_11459_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1073_cast_fp16)[name = string("op_11459_cast_fp16")]; tensor var_11460_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1075_cast_fp16)[name = string("op_11460_cast_fp16")]; tensor var_11461_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1077_cast_fp16)[name = string("op_11461_cast_fp16")]; tensor var_11462_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1079_cast_fp16)[name = string("op_11462_cast_fp16")]; tensor var_11463_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1081_cast_fp16)[name = string("op_11463_cast_fp16")]; tensor var_11464_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1083_cast_fp16)[name = string("op_11464_cast_fp16")]; tensor var_11465_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1085_cast_fp16)[name = string("op_11465_cast_fp16")]; tensor var_11466_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1087_cast_fp16)[name = string("op_11466_cast_fp16")]; tensor var_11467_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1089_cast_fp16)[name = string("op_11467_cast_fp16")]; tensor var_11468_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1091_cast_fp16)[name = string("op_11468_cast_fp16")]; tensor var_11469_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1093_cast_fp16)[name = string("op_11469_cast_fp16")]; tensor var_11470_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1095_cast_fp16)[name = string("op_11470_cast_fp16")]; tensor var_11471_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1097_cast_fp16)[name = string("op_11471_cast_fp16")]; tensor var_11472_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1099_cast_fp16)[name = string("op_11472_cast_fp16")]; tensor var_11473_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1101_cast_fp16)[name = string("op_11473_cast_fp16")]; tensor var_11474_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1103_cast_fp16)[name = string("op_11474_cast_fp16")]; tensor var_11475_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1105_cast_fp16)[name = string("op_11475_cast_fp16")]; tensor var_11476_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1107_cast_fp16)[name = string("op_11476_cast_fp16")]; tensor var_11477_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1109_cast_fp16)[name = string("op_11477_cast_fp16")]; tensor var_11478_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1111_cast_fp16)[name = string("op_11478_cast_fp16")]; tensor var_11479_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1113_cast_fp16)[name = string("op_11479_cast_fp16")]; tensor var_11480_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1115_cast_fp16)[name = string("op_11480_cast_fp16")]; tensor var_11481_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1117_cast_fp16)[name = string("op_11481_cast_fp16")]; tensor var_11482_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1119_cast_fp16)[name = string("op_11482_cast_fp16")]; tensor var_11483_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1121_cast_fp16)[name = string("op_11483_cast_fp16")]; tensor var_11484_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1123_cast_fp16)[name = string("op_11484_cast_fp16")]; tensor var_11485_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1125_cast_fp16)[name = string("op_11485_cast_fp16")]; tensor var_11486_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1127_cast_fp16)[name = string("op_11486_cast_fp16")]; tensor var_11487_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1129_cast_fp16)[name = string("op_11487_cast_fp16")]; tensor var_11488_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1131_cast_fp16)[name = string("op_11488_cast_fp16")]; tensor var_11489_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1133_cast_fp16)[name = string("op_11489_cast_fp16")]; tensor var_11490_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1135_cast_fp16)[name = string("op_11490_cast_fp16")]; tensor var_11491_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1137_cast_fp16)[name = string("op_11491_cast_fp16")]; tensor var_11492_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1139_cast_fp16)[name = string("op_11492_cast_fp16")]; tensor var_11493_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1141_cast_fp16)[name = string("op_11493_cast_fp16")]; tensor var_11494_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1143_cast_fp16)[name = string("op_11494_cast_fp16")]; tensor var_11495_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1145_cast_fp16)[name = string("op_11495_cast_fp16")]; tensor var_11496_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1147_cast_fp16)[name = string("op_11496_cast_fp16")]; tensor var_11497_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1149_cast_fp16)[name = string("op_11497_cast_fp16")]; tensor var_11498_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_cast_fp16)[name = string("op_11498_cast_fp16")]; string var_11500_equation_0 = const()[name = string("op_11500_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11500_cast_fp16 = einsum(equation = var_11500_equation_0, values = (var_11212_cast_fp16, var_11451_cast_fp16))[name = string("op_11500_cast_fp16")]; string var_11502_equation_0 = const()[name = string("op_11502_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11502_cast_fp16 = einsum(equation = var_11502_equation_0, values = (var_11212_cast_fp16, var_11452_cast_fp16))[name = string("op_11502_cast_fp16")]; string var_11504_equation_0 = const()[name = string("op_11504_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11504_cast_fp16 = einsum(equation = var_11504_equation_0, values = (var_11212_cast_fp16, var_11453_cast_fp16))[name = string("op_11504_cast_fp16")]; string var_11506_equation_0 = const()[name = string("op_11506_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11506_cast_fp16 = einsum(equation = var_11506_equation_0, values = (var_11212_cast_fp16, var_11454_cast_fp16))[name = string("op_11506_cast_fp16")]; string var_11508_equation_0 = const()[name = string("op_11508_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11508_cast_fp16 = einsum(equation = var_11508_equation_0, values = (var_11216_cast_fp16, var_11455_cast_fp16))[name = string("op_11508_cast_fp16")]; string var_11510_equation_0 = const()[name = string("op_11510_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11510_cast_fp16 = einsum(equation = var_11510_equation_0, values = (var_11216_cast_fp16, var_11456_cast_fp16))[name = string("op_11510_cast_fp16")]; string var_11512_equation_0 = const()[name = string("op_11512_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11512_cast_fp16 = einsum(equation = var_11512_equation_0, values = (var_11216_cast_fp16, var_11457_cast_fp16))[name = string("op_11512_cast_fp16")]; string var_11514_equation_0 = const()[name = string("op_11514_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11514_cast_fp16 = einsum(equation = var_11514_equation_0, values = (var_11216_cast_fp16, var_11458_cast_fp16))[name = string("op_11514_cast_fp16")]; string var_11516_equation_0 = const()[name = string("op_11516_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11516_cast_fp16 = einsum(equation = var_11516_equation_0, values = (var_11220_cast_fp16, var_11459_cast_fp16))[name = string("op_11516_cast_fp16")]; string var_11518_equation_0 = const()[name = string("op_11518_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11518_cast_fp16 = einsum(equation = var_11518_equation_0, values = (var_11220_cast_fp16, var_11460_cast_fp16))[name = string("op_11518_cast_fp16")]; string var_11520_equation_0 = const()[name = string("op_11520_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11520_cast_fp16 = einsum(equation = var_11520_equation_0, values = (var_11220_cast_fp16, var_11461_cast_fp16))[name = string("op_11520_cast_fp16")]; string var_11522_equation_0 = const()[name = string("op_11522_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11522_cast_fp16 = einsum(equation = var_11522_equation_0, values = (var_11220_cast_fp16, var_11462_cast_fp16))[name = string("op_11522_cast_fp16")]; string var_11524_equation_0 = const()[name = string("op_11524_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11524_cast_fp16 = einsum(equation = var_11524_equation_0, values = (var_11224_cast_fp16, var_11463_cast_fp16))[name = string("op_11524_cast_fp16")]; string var_11526_equation_0 = const()[name = string("op_11526_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11526_cast_fp16 = einsum(equation = var_11526_equation_0, values = (var_11224_cast_fp16, var_11464_cast_fp16))[name = string("op_11526_cast_fp16")]; string var_11528_equation_0 = const()[name = string("op_11528_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11528_cast_fp16 = einsum(equation = var_11528_equation_0, values = (var_11224_cast_fp16, var_11465_cast_fp16))[name = string("op_11528_cast_fp16")]; string var_11530_equation_0 = const()[name = string("op_11530_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11530_cast_fp16 = einsum(equation = var_11530_equation_0, values = (var_11224_cast_fp16, var_11466_cast_fp16))[name = string("op_11530_cast_fp16")]; string var_11532_equation_0 = const()[name = string("op_11532_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11532_cast_fp16 = einsum(equation = var_11532_equation_0, values = (var_11228_cast_fp16, var_11467_cast_fp16))[name = string("op_11532_cast_fp16")]; string var_11534_equation_0 = const()[name = string("op_11534_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11534_cast_fp16 = einsum(equation = var_11534_equation_0, values = (var_11228_cast_fp16, var_11468_cast_fp16))[name = string("op_11534_cast_fp16")]; string var_11536_equation_0 = const()[name = string("op_11536_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11536_cast_fp16 = einsum(equation = var_11536_equation_0, values = (var_11228_cast_fp16, var_11469_cast_fp16))[name = string("op_11536_cast_fp16")]; string var_11538_equation_0 = const()[name = string("op_11538_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11538_cast_fp16 = einsum(equation = var_11538_equation_0, values = (var_11228_cast_fp16, var_11470_cast_fp16))[name = string("op_11538_cast_fp16")]; string var_11540_equation_0 = const()[name = string("op_11540_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11540_cast_fp16 = einsum(equation = var_11540_equation_0, values = (var_11232_cast_fp16, var_11471_cast_fp16))[name = string("op_11540_cast_fp16")]; string var_11542_equation_0 = const()[name = string("op_11542_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11542_cast_fp16 = einsum(equation = var_11542_equation_0, values = (var_11232_cast_fp16, var_11472_cast_fp16))[name = string("op_11542_cast_fp16")]; string var_11544_equation_0 = const()[name = string("op_11544_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11544_cast_fp16 = einsum(equation = var_11544_equation_0, values = (var_11232_cast_fp16, var_11473_cast_fp16))[name = string("op_11544_cast_fp16")]; string var_11546_equation_0 = const()[name = string("op_11546_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11546_cast_fp16 = einsum(equation = var_11546_equation_0, values = (var_11232_cast_fp16, var_11474_cast_fp16))[name = string("op_11546_cast_fp16")]; string var_11548_equation_0 = const()[name = string("op_11548_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11236_cast_fp16, var_11475_cast_fp16))[name = string("op_11548_cast_fp16")]; string var_11550_equation_0 = const()[name = string("op_11550_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11550_cast_fp16 = einsum(equation = var_11550_equation_0, values = (var_11236_cast_fp16, var_11476_cast_fp16))[name = string("op_11550_cast_fp16")]; string var_11552_equation_0 = const()[name = string("op_11552_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11236_cast_fp16, var_11477_cast_fp16))[name = string("op_11552_cast_fp16")]; string var_11554_equation_0 = const()[name = string("op_11554_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11554_cast_fp16 = einsum(equation = var_11554_equation_0, values = (var_11236_cast_fp16, var_11478_cast_fp16))[name = string("op_11554_cast_fp16")]; string var_11556_equation_0 = const()[name = string("op_11556_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11240_cast_fp16, var_11479_cast_fp16))[name = string("op_11556_cast_fp16")]; string var_11558_equation_0 = const()[name = string("op_11558_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11558_cast_fp16 = einsum(equation = var_11558_equation_0, values = (var_11240_cast_fp16, var_11480_cast_fp16))[name = string("op_11558_cast_fp16")]; string var_11560_equation_0 = const()[name = string("op_11560_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11240_cast_fp16, var_11481_cast_fp16))[name = string("op_11560_cast_fp16")]; string var_11562_equation_0 = const()[name = string("op_11562_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11562_cast_fp16 = einsum(equation = var_11562_equation_0, values = (var_11240_cast_fp16, var_11482_cast_fp16))[name = string("op_11562_cast_fp16")]; string var_11564_equation_0 = const()[name = string("op_11564_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11244_cast_fp16, var_11483_cast_fp16))[name = string("op_11564_cast_fp16")]; string var_11566_equation_0 = const()[name = string("op_11566_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11566_cast_fp16 = einsum(equation = var_11566_equation_0, values = (var_11244_cast_fp16, var_11484_cast_fp16))[name = string("op_11566_cast_fp16")]; string var_11568_equation_0 = const()[name = string("op_11568_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11568_cast_fp16 = einsum(equation = var_11568_equation_0, values = (var_11244_cast_fp16, var_11485_cast_fp16))[name = string("op_11568_cast_fp16")]; string var_11570_equation_0 = const()[name = string("op_11570_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11570_cast_fp16 = einsum(equation = var_11570_equation_0, values = (var_11244_cast_fp16, var_11486_cast_fp16))[name = string("op_11570_cast_fp16")]; string var_11572_equation_0 = const()[name = string("op_11572_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11572_cast_fp16 = einsum(equation = var_11572_equation_0, values = (var_11248_cast_fp16, var_11487_cast_fp16))[name = string("op_11572_cast_fp16")]; string var_11574_equation_0 = const()[name = string("op_11574_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11574_cast_fp16 = einsum(equation = var_11574_equation_0, values = (var_11248_cast_fp16, var_11488_cast_fp16))[name = string("op_11574_cast_fp16")]; string var_11576_equation_0 = const()[name = string("op_11576_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11576_cast_fp16 = einsum(equation = var_11576_equation_0, values = (var_11248_cast_fp16, var_11489_cast_fp16))[name = string("op_11576_cast_fp16")]; string var_11578_equation_0 = const()[name = string("op_11578_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11578_cast_fp16 = einsum(equation = var_11578_equation_0, values = (var_11248_cast_fp16, var_11490_cast_fp16))[name = string("op_11578_cast_fp16")]; string var_11580_equation_0 = const()[name = string("op_11580_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11580_cast_fp16 = einsum(equation = var_11580_equation_0, values = (var_11252_cast_fp16, var_11491_cast_fp16))[name = string("op_11580_cast_fp16")]; string var_11582_equation_0 = const()[name = string("op_11582_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11582_cast_fp16 = einsum(equation = var_11582_equation_0, values = (var_11252_cast_fp16, var_11492_cast_fp16))[name = string("op_11582_cast_fp16")]; string var_11584_equation_0 = const()[name = string("op_11584_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11584_cast_fp16 = einsum(equation = var_11584_equation_0, values = (var_11252_cast_fp16, var_11493_cast_fp16))[name = string("op_11584_cast_fp16")]; string var_11586_equation_0 = const()[name = string("op_11586_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11586_cast_fp16 = einsum(equation = var_11586_equation_0, values = (var_11252_cast_fp16, var_11494_cast_fp16))[name = string("op_11586_cast_fp16")]; string var_11588_equation_0 = const()[name = string("op_11588_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11588_cast_fp16 = einsum(equation = var_11588_equation_0, values = (var_11256_cast_fp16, var_11495_cast_fp16))[name = string("op_11588_cast_fp16")]; string var_11590_equation_0 = const()[name = string("op_11590_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11590_cast_fp16 = einsum(equation = var_11590_equation_0, values = (var_11256_cast_fp16, var_11496_cast_fp16))[name = string("op_11590_cast_fp16")]; string var_11592_equation_0 = const()[name = string("op_11592_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11592_cast_fp16 = einsum(equation = var_11592_equation_0, values = (var_11256_cast_fp16, var_11497_cast_fp16))[name = string("op_11592_cast_fp16")]; string var_11594_equation_0 = const()[name = string("op_11594_equation_0"), val = string("bchk,bkhq->bchq")]; tensor var_11594_cast_fp16 = einsum(equation = var_11594_equation_0, values = (var_11256_cast_fp16, var_11498_cast_fp16))[name = string("op_11594_cast_fp16")]; bool var_11596_interleave_0 = const()[name = string("op_11596_interleave_0"), val = bool(false)]; tensor var_11596_cast_fp16 = concat(axis = var_10707, interleave = var_11596_interleave_0, values = (var_11500_cast_fp16, var_11502_cast_fp16, var_11504_cast_fp16, var_11506_cast_fp16))[name = string("op_11596_cast_fp16")]; bool var_11598_interleave_0 = const()[name = string("op_11598_interleave_0"), val = bool(false)]; tensor var_11598_cast_fp16 = concat(axis = var_10707, interleave = var_11598_interleave_0, values = (var_11508_cast_fp16, var_11510_cast_fp16, var_11512_cast_fp16, var_11514_cast_fp16))[name = string("op_11598_cast_fp16")]; bool var_11600_interleave_0 = const()[name = string("op_11600_interleave_0"), val = bool(false)]; tensor var_11600_cast_fp16 = concat(axis = var_10707, interleave = var_11600_interleave_0, values = (var_11516_cast_fp16, var_11518_cast_fp16, var_11520_cast_fp16, var_11522_cast_fp16))[name = string("op_11600_cast_fp16")]; bool var_11602_interleave_0 = const()[name = string("op_11602_interleave_0"), val = bool(false)]; tensor var_11602_cast_fp16 = concat(axis = var_10707, interleave = var_11602_interleave_0, values = (var_11524_cast_fp16, var_11526_cast_fp16, var_11528_cast_fp16, var_11530_cast_fp16))[name = string("op_11602_cast_fp16")]; bool var_11604_interleave_0 = const()[name = string("op_11604_interleave_0"), val = bool(false)]; tensor var_11604_cast_fp16 = concat(axis = var_10707, interleave = var_11604_interleave_0, values = (var_11532_cast_fp16, var_11534_cast_fp16, var_11536_cast_fp16, var_11538_cast_fp16))[name = string("op_11604_cast_fp16")]; bool var_11606_interleave_0 = const()[name = string("op_11606_interleave_0"), val = bool(false)]; tensor var_11606_cast_fp16 = concat(axis = var_10707, interleave = var_11606_interleave_0, values = (var_11540_cast_fp16, var_11542_cast_fp16, var_11544_cast_fp16, var_11546_cast_fp16))[name = string("op_11606_cast_fp16")]; bool var_11608_interleave_0 = const()[name = string("op_11608_interleave_0"), val = bool(false)]; tensor var_11608_cast_fp16 = concat(axis = var_10707, interleave = var_11608_interleave_0, values = (var_11548_cast_fp16, var_11550_cast_fp16, var_11552_cast_fp16, var_11554_cast_fp16))[name = string("op_11608_cast_fp16")]; bool var_11610_interleave_0 = const()[name = string("op_11610_interleave_0"), val = bool(false)]; tensor var_11610_cast_fp16 = concat(axis = var_10707, interleave = var_11610_interleave_0, values = (var_11556_cast_fp16, var_11558_cast_fp16, var_11560_cast_fp16, var_11562_cast_fp16))[name = string("op_11610_cast_fp16")]; bool var_11612_interleave_0 = const()[name = string("op_11612_interleave_0"), val = bool(false)]; tensor var_11612_cast_fp16 = concat(axis = var_10707, interleave = var_11612_interleave_0, values = (var_11564_cast_fp16, var_11566_cast_fp16, var_11568_cast_fp16, var_11570_cast_fp16))[name = string("op_11612_cast_fp16")]; bool var_11614_interleave_0 = const()[name = string("op_11614_interleave_0"), val = bool(false)]; tensor var_11614_cast_fp16 = concat(axis = var_10707, interleave = var_11614_interleave_0, values = (var_11572_cast_fp16, var_11574_cast_fp16, var_11576_cast_fp16, var_11578_cast_fp16))[name = string("op_11614_cast_fp16")]; bool var_11616_interleave_0 = const()[name = string("op_11616_interleave_0"), val = bool(false)]; tensor var_11616_cast_fp16 = concat(axis = var_10707, interleave = var_11616_interleave_0, values = (var_11580_cast_fp16, var_11582_cast_fp16, var_11584_cast_fp16, var_11586_cast_fp16))[name = string("op_11616_cast_fp16")]; bool var_11618_interleave_0 = const()[name = string("op_11618_interleave_0"), val = bool(false)]; tensor var_11618_cast_fp16 = concat(axis = var_10707, interleave = var_11618_interleave_0, values = (var_11588_cast_fp16, var_11590_cast_fp16, var_11592_cast_fp16, var_11594_cast_fp16))[name = string("op_11618_cast_fp16")]; bool input_89_interleave_0 = const()[name = string("input_89_interleave_0"), val = bool(false)]; tensor input_89_cast_fp16 = concat(axis = var_10724, interleave = input_89_interleave_0, values = (var_11596_cast_fp16, var_11598_cast_fp16, var_11600_cast_fp16, var_11602_cast_fp16, var_11604_cast_fp16, var_11606_cast_fp16, var_11608_cast_fp16, var_11610_cast_fp16, var_11612_cast_fp16, var_11614_cast_fp16, var_11616_cast_fp16, var_11618_cast_fp16))[name = string("input_89_cast_fp16")]; string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165690624)))]; tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166870336)))]; tensor obj_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_cast_fp16)[name = string("inputs_47_cast_fp16")]; tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; fp16 var_11637_to_fp16 = const()[name = string("op_11637_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_11637_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166871936)))]; tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166873536)))]; fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166875136)))]; tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171593792)))]; tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171600000)))]; tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176318656)))]; tensor hidden_states_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_cast_fp16")]; tensor inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; fp16 var_11675_to_fp16 = const()[name = string("op_11675_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_11675_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176320256)))]; tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176321856)))]; fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; string var_11699_pad_type_0 = const()[name = string("op_11699_pad_type_0"), val = string("valid")]; tensor var_11699_strides_0 = const()[name = string("op_11699_strides_0"), val = tensor([1, 1])]; tensor var_11699_pad_0 = const()[name = string("op_11699_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11699_dilations_0 = const()[name = string("op_11699_dilations_0"), val = tensor([1, 1])]; int32 var_11699_groups_0 = const()[name = string("op_11699_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176323456)))]; tensor var_11699_cast_fp16 = conv(dilations = var_11699_dilations_0, groups = var_11699_groups_0, pad = var_11699_pad_0, pad_type = var_11699_pad_type_0, strides = var_11699_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11699_cast_fp16")]; string var_11706_pad_type_0 = const()[name = string("op_11706_pad_type_0"), val = string("valid")]; tensor var_11706_strides_0 = const()[name = string("op_11706_strides_0"), val = tensor([1, 1])]; tensor var_11706_pad_0 = const()[name = string("op_11706_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11706_dilations_0 = const()[name = string("op_11706_dilations_0"), val = tensor([1, 1])]; int32 var_11706_groups_0 = const()[name = string("op_11706_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177503168)))]; tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178682880)))]; tensor var_11706_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_11706_dilations_0, groups = var_11706_groups_0, pad = var_11706_pad_0, pad_type = var_11706_pad_type_0, strides = var_11706_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11706_cast_fp16")]; string var_11724_pad_type_0 = const()[name = string("op_11724_pad_type_0"), val = string("valid")]; tensor var_11724_strides_0 = const()[name = string("op_11724_strides_0"), val = tensor([1, 1])]; tensor var_11724_pad_0 = const()[name = string("op_11724_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11724_dilations_0 = const()[name = string("op_11724_dilations_0"), val = tensor([1, 1])]; int32 var_11724_groups_0 = const()[name = string("op_11724_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178684480)))]; tensor var_11724_cast_fp16 = conv(dilations = var_11724_dilations_0, groups = var_11724_groups_0, pad = var_11724_pad_0, pad_type = var_11724_pad_type_0, strides = var_11724_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11724_cast_fp16")]; string var_11731_pad_type_0 = const()[name = string("op_11731_pad_type_0"), val = string("valid")]; tensor var_11731_strides_0 = const()[name = string("op_11731_strides_0"), val = tensor([1, 1])]; tensor var_11731_pad_0 = const()[name = string("op_11731_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11731_dilations_0 = const()[name = string("op_11731_dilations_0"), val = tensor([1, 1])]; int32 var_11731_groups_0 = const()[name = string("op_11731_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179864192)))]; tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181043904)))]; tensor var_11731_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_11731_dilations_0, groups = var_11731_groups_0, pad = var_11731_pad_0, pad_type = var_11731_pad_type_0, strides = var_11731_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11731_cast_fp16")]; string var_11749_pad_type_0 = const()[name = string("op_11749_pad_type_0"), val = string("valid")]; tensor var_11749_strides_0 = const()[name = string("op_11749_strides_0"), val = tensor([1, 1])]; tensor var_11749_pad_0 = const()[name = string("op_11749_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11749_dilations_0 = const()[name = string("op_11749_dilations_0"), val = tensor([1, 1])]; int32 var_11749_groups_0 = const()[name = string("op_11749_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181045504)))]; tensor var_11749_cast_fp16 = conv(dilations = var_11749_dilations_0, groups = var_11749_groups_0, pad = var_11749_pad_0, pad_type = var_11749_pad_type_0, strides = var_11749_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11749_cast_fp16")]; string var_11756_pad_type_0 = const()[name = string("op_11756_pad_type_0"), val = string("valid")]; tensor var_11756_strides_0 = const()[name = string("op_11756_strides_0"), val = tensor([1, 1])]; tensor var_11756_pad_0 = const()[name = string("op_11756_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11756_dilations_0 = const()[name = string("op_11756_dilations_0"), val = tensor([1, 1])]; int32 var_11756_groups_0 = const()[name = string("op_11756_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182225216)))]; tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183404928)))]; tensor var_11756_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_11756_dilations_0, groups = var_11756_groups_0, pad = var_11756_pad_0, pad_type = var_11756_pad_type_0, strides = var_11756_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11756_cast_fp16")]; string var_11774_pad_type_0 = const()[name = string("op_11774_pad_type_0"), val = string("valid")]; tensor var_11774_strides_0 = const()[name = string("op_11774_strides_0"), val = tensor([1, 1])]; tensor var_11774_pad_0 = const()[name = string("op_11774_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11774_dilations_0 = const()[name = string("op_11774_dilations_0"), val = tensor([1, 1])]; int32 var_11774_groups_0 = const()[name = string("op_11774_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183406528)))]; tensor var_11774_cast_fp16 = conv(dilations = var_11774_dilations_0, groups = var_11774_groups_0, pad = var_11774_pad_0, pad_type = var_11774_pad_type_0, strides = var_11774_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11774_cast_fp16")]; string var_11781_pad_type_0 = const()[name = string("op_11781_pad_type_0"), val = string("valid")]; tensor var_11781_strides_0 = const()[name = string("op_11781_strides_0"), val = tensor([1, 1])]; tensor var_11781_pad_0 = const()[name = string("op_11781_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11781_dilations_0 = const()[name = string("op_11781_dilations_0"), val = tensor([1, 1])]; int32 var_11781_groups_0 = const()[name = string("op_11781_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184586240)))]; tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185765952)))]; tensor var_11781_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_11781_dilations_0, groups = var_11781_groups_0, pad = var_11781_pad_0, pad_type = var_11781_pad_type_0, strides = var_11781_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11781_cast_fp16")]; string var_11799_pad_type_0 = const()[name = string("op_11799_pad_type_0"), val = string("valid")]; tensor var_11799_strides_0 = const()[name = string("op_11799_strides_0"), val = tensor([1, 1])]; tensor var_11799_pad_0 = const()[name = string("op_11799_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11799_dilations_0 = const()[name = string("op_11799_dilations_0"), val = tensor([1, 1])]; int32 var_11799_groups_0 = const()[name = string("op_11799_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185767552)))]; tensor var_11799_cast_fp16 = conv(dilations = var_11799_dilations_0, groups = var_11799_groups_0, pad = var_11799_pad_0, pad_type = var_11799_pad_type_0, strides = var_11799_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11799_cast_fp16")]; string var_11806_pad_type_0 = const()[name = string("op_11806_pad_type_0"), val = string("valid")]; tensor var_11806_strides_0 = const()[name = string("op_11806_strides_0"), val = tensor([1, 1])]; tensor var_11806_pad_0 = const()[name = string("op_11806_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11806_dilations_0 = const()[name = string("op_11806_dilations_0"), val = tensor([1, 1])]; int32 var_11806_groups_0 = const()[name = string("op_11806_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186947264)))]; tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188126976)))]; tensor var_11806_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_11806_dilations_0, groups = var_11806_groups_0, pad = var_11806_pad_0, pad_type = var_11806_pad_type_0, strides = var_11806_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11806_cast_fp16")]; string var_11824_pad_type_0 = const()[name = string("op_11824_pad_type_0"), val = string("valid")]; tensor var_11824_strides_0 = const()[name = string("op_11824_strides_0"), val = tensor([1, 1])]; tensor var_11824_pad_0 = const()[name = string("op_11824_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11824_dilations_0 = const()[name = string("op_11824_dilations_0"), val = tensor([1, 1])]; int32 var_11824_groups_0 = const()[name = string("op_11824_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188128576)))]; tensor var_11824_cast_fp16 = conv(dilations = var_11824_dilations_0, groups = var_11824_groups_0, pad = var_11824_pad_0, pad_type = var_11824_pad_type_0, strides = var_11824_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11824_cast_fp16")]; string var_11831_pad_type_0 = const()[name = string("op_11831_pad_type_0"), val = string("valid")]; tensor var_11831_strides_0 = const()[name = string("op_11831_strides_0"), val = tensor([1, 1])]; tensor var_11831_pad_0 = const()[name = string("op_11831_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11831_dilations_0 = const()[name = string("op_11831_dilations_0"), val = tensor([1, 1])]; int32 var_11831_groups_0 = const()[name = string("op_11831_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189308288)))]; tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190488000)))]; tensor var_11831_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = var_11831_dilations_0, groups = var_11831_groups_0, pad = var_11831_pad_0, pad_type = var_11831_pad_type_0, strides = var_11831_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11831_cast_fp16")]; string var_11849_pad_type_0 = const()[name = string("op_11849_pad_type_0"), val = string("valid")]; tensor var_11849_strides_0 = const()[name = string("op_11849_strides_0"), val = tensor([1, 1])]; tensor var_11849_pad_0 = const()[name = string("op_11849_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11849_dilations_0 = const()[name = string("op_11849_dilations_0"), val = tensor([1, 1])]; int32 var_11849_groups_0 = const()[name = string("op_11849_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190489600)))]; tensor var_11849_cast_fp16 = conv(dilations = var_11849_dilations_0, groups = var_11849_groups_0, pad = var_11849_pad_0, pad_type = var_11849_pad_type_0, strides = var_11849_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11849_cast_fp16")]; string var_11856_pad_type_0 = const()[name = string("op_11856_pad_type_0"), val = string("valid")]; tensor var_11856_strides_0 = const()[name = string("op_11856_strides_0"), val = tensor([1, 1])]; tensor var_11856_pad_0 = const()[name = string("op_11856_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11856_dilations_0 = const()[name = string("op_11856_dilations_0"), val = tensor([1, 1])]; int32 var_11856_groups_0 = const()[name = string("op_11856_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191669312)))]; tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192849024)))]; tensor var_11856_cast_fp16 = conv(bias = decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16, dilations = var_11856_dilations_0, groups = var_11856_groups_0, pad = var_11856_pad_0, pad_type = var_11856_pad_type_0, strides = var_11856_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11856_cast_fp16")]; string var_11874_pad_type_0 = const()[name = string("op_11874_pad_type_0"), val = string("valid")]; tensor var_11874_strides_0 = const()[name = string("op_11874_strides_0"), val = tensor([1, 1])]; tensor var_11874_pad_0 = const()[name = string("op_11874_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11874_dilations_0 = const()[name = string("op_11874_dilations_0"), val = tensor([1, 1])]; int32 var_11874_groups_0 = const()[name = string("op_11874_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192850624)))]; tensor var_11874_cast_fp16 = conv(dilations = var_11874_dilations_0, groups = var_11874_groups_0, pad = var_11874_pad_0, pad_type = var_11874_pad_type_0, strides = var_11874_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11874_cast_fp16")]; string var_11881_pad_type_0 = const()[name = string("op_11881_pad_type_0"), val = string("valid")]; tensor var_11881_strides_0 = const()[name = string("op_11881_strides_0"), val = tensor([1, 1])]; tensor var_11881_pad_0 = const()[name = string("op_11881_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11881_dilations_0 = const()[name = string("op_11881_dilations_0"), val = tensor([1, 1])]; int32 var_11881_groups_0 = const()[name = string("op_11881_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194030336)))]; tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195210048)))]; tensor var_11881_cast_fp16 = conv(bias = decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16, dilations = var_11881_dilations_0, groups = var_11881_groups_0, pad = var_11881_pad_0, pad_type = var_11881_pad_type_0, strides = var_11881_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11881_cast_fp16")]; string var_11899_pad_type_0 = const()[name = string("op_11899_pad_type_0"), val = string("valid")]; tensor var_11899_strides_0 = const()[name = string("op_11899_strides_0"), val = tensor([1, 1])]; tensor var_11899_pad_0 = const()[name = string("op_11899_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11899_dilations_0 = const()[name = string("op_11899_dilations_0"), val = tensor([1, 1])]; int32 var_11899_groups_0 = const()[name = string("op_11899_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195211648)))]; tensor var_11899_cast_fp16 = conv(dilations = var_11899_dilations_0, groups = var_11899_groups_0, pad = var_11899_pad_0, pad_type = var_11899_pad_type_0, strides = var_11899_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11899_cast_fp16")]; string var_11906_pad_type_0 = const()[name = string("op_11906_pad_type_0"), val = string("valid")]; tensor var_11906_strides_0 = const()[name = string("op_11906_strides_0"), val = tensor([1, 1])]; tensor var_11906_pad_0 = const()[name = string("op_11906_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11906_dilations_0 = const()[name = string("op_11906_dilations_0"), val = tensor([1, 1])]; int32 var_11906_groups_0 = const()[name = string("op_11906_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196391360)))]; tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197571072)))]; tensor var_11906_cast_fp16 = conv(bias = decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16, dilations = var_11906_dilations_0, groups = var_11906_groups_0, pad = var_11906_pad_0, pad_type = var_11906_pad_type_0, strides = var_11906_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11906_cast_fp16")]; string var_11924_pad_type_0 = const()[name = string("op_11924_pad_type_0"), val = string("valid")]; tensor var_11924_strides_0 = const()[name = string("op_11924_strides_0"), val = tensor([1, 1])]; tensor var_11924_pad_0 = const()[name = string("op_11924_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11924_dilations_0 = const()[name = string("op_11924_dilations_0"), val = tensor([1, 1])]; int32 var_11924_groups_0 = const()[name = string("op_11924_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197572672)))]; tensor var_11924_cast_fp16 = conv(dilations = var_11924_dilations_0, groups = var_11924_groups_0, pad = var_11924_pad_0, pad_type = var_11924_pad_type_0, strides = var_11924_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11924_cast_fp16")]; string var_11931_pad_type_0 = const()[name = string("op_11931_pad_type_0"), val = string("valid")]; tensor var_11931_strides_0 = const()[name = string("op_11931_strides_0"), val = tensor([1, 1])]; tensor var_11931_pad_0 = const()[name = string("op_11931_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11931_dilations_0 = const()[name = string("op_11931_dilations_0"), val = tensor([1, 1])]; int32 var_11931_groups_0 = const()[name = string("op_11931_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198752384)))]; tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199932096)))]; tensor var_11931_cast_fp16 = conv(bias = decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16, dilations = var_11931_dilations_0, groups = var_11931_groups_0, pad = var_11931_pad_0, pad_type = var_11931_pad_type_0, strides = var_11931_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11931_cast_fp16")]; string var_11949_pad_type_0 = const()[name = string("op_11949_pad_type_0"), val = string("valid")]; tensor var_11949_strides_0 = const()[name = string("op_11949_strides_0"), val = tensor([1, 1])]; tensor var_11949_pad_0 = const()[name = string("op_11949_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11949_dilations_0 = const()[name = string("op_11949_dilations_0"), val = tensor([1, 1])]; int32 var_11949_groups_0 = const()[name = string("op_11949_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199933696)))]; tensor var_11949_cast_fp16 = conv(dilations = var_11949_dilations_0, groups = var_11949_groups_0, pad = var_11949_pad_0, pad_type = var_11949_pad_type_0, strides = var_11949_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11949_cast_fp16")]; string var_11956_pad_type_0 = const()[name = string("op_11956_pad_type_0"), val = string("valid")]; tensor var_11956_strides_0 = const()[name = string("op_11956_strides_0"), val = tensor([1, 1])]; tensor var_11956_pad_0 = const()[name = string("op_11956_pad_0"), val = tensor([0, 0, 0, 0])]; tensor var_11956_dilations_0 = const()[name = string("op_11956_dilations_0"), val = tensor([1, 1])]; int32 var_11956_groups_0 = const()[name = string("op_11956_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201113408)))]; tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202293120)))]; tensor var_11956_cast_fp16 = conv(bias = decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16, dilations = var_11956_dilations_0, groups = var_11956_groups_0, pad = var_11956_pad_0, pad_type = var_11956_pad_type_0, strides = var_11956_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11956_cast_fp16")]; string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202294720)))]; tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203474432)))]; tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204654144)))]; tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; int32 var_11986 = const()[name = string("op_11986"), val = int32(0)]; bool input_99_interleave_0 = const()[name = string("input_99_interleave_0"), val = bool(false)]; tensor input_99_cast_fp16 = concat(axis = var_11986, interleave = input_99_interleave_0, values = (var_11699_cast_fp16, var_11724_cast_fp16, var_11749_cast_fp16, var_11774_cast_fp16, var_11799_cast_fp16, var_11824_cast_fp16, var_11849_cast_fp16, var_11874_cast_fp16, var_11899_cast_fp16, var_11924_cast_fp16, var_11949_cast_fp16, k_cast_fp16))[name = string("input_99_cast_fp16")]; int32 var_11989 = const()[name = string("op_11989"), val = int32(0)]; bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; tensor input_cast_fp16 = concat(axis = var_11989, interleave = input_interleave_0, values = (var_11706_cast_fp16, var_11731_cast_fp16, var_11756_cast_fp16, var_11781_cast_fp16, var_11806_cast_fp16, var_11831_cast_fp16, var_11856_cast_fp16, var_11881_cast_fp16, var_11906_cast_fp16, var_11931_cast_fp16, var_11956_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; tensor var_11996_pad_0 = const()[name = string("op_11996_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; string var_11996_mode_0 = const()[name = string("op_11996_mode_0"), val = string("constant")]; fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)]; tensor encoder_attn_key_cache = pad(constant_val = const_13_to_fp16, mode = var_11996_mode_0, pad = var_11996_pad_0, x = input_99_cast_fp16)[name = string("op_11996_cast_fp16")]; tensor var_12002_pad_0 = const()[name = string("op_12002_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; string var_12002_mode_0 = const()[name = string("op_12002_mode_0"), val = string("constant")]; fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)]; tensor encoder_attn_value_cache = pad(constant_val = const_14_to_fp16, mode = var_12002_mode_0, pad = var_12002_pad_0, x = input_cast_fp16)[name = string("op_12002_cast_fp16")]; } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); }