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 cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { int32 var_42_axis_0 = const()[name = string("op_42_axis_0"), val = int32(0)]; int32 var_42_batch_dims_0 = const()[name = string("op_42_batch_dims_0"), val = int32(0)]; bool var_42_validate_indices_0 = const()[name = string("op_42_validate_indices_0"), val = bool(false)]; tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor var_42_cast_fp16 = gather(axis = var_42_axis_0, batch_dims = var_42_batch_dims_0, indices = input_ids, validate_indices = var_42_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_42_cast_fp16")]; int32 var_46_axis_0 = const()[name = string("op_46_axis_0"), val = int32(0)]; int32 var_46_batch_dims_0 = const()[name = string("op_46_batch_dims_0"), val = int32(0)]; bool var_46_validate_indices_0 = const()[name = string("op_46_validate_indices_0"), val = bool(false)]; tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79663232)))]; string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_183")]; tensor var_46_cast_fp16_cast_uint16 = gather(axis = var_46_axis_0, batch_dims = var_46_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_46_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_46_cast_fp16_cast_uint16")]; tensor hidden_states_1_cast_fp16 = add(x = var_42_cast_fp16, y = var_46_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; tensor var_60_axes_0 = const()[name = string("op_60_axes_0"), val = tensor([2])]; tensor var_60_cast_fp16 = expand_dims(axes = var_60_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_60_cast_fp16")]; tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_60_cast_fp16)[name = string("inputs_1_cast_fp16")]; tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; tensor tile_0 = const()[name = string("tile_0"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351424)))]; int32 var_65_axis_0 = const()[name = string("op_65_axis_0"), val = int32(0)]; tensor var_65_cast_fp16_0, tensor var_65_cast_fp16_1, tensor var_65_cast_fp16_2, tensor var_65_cast_fp16_3, tensor var_65_cast_fp16_4, tensor var_65_cast_fp16_5, tensor var_65_cast_fp16_6, tensor var_65_cast_fp16_7, tensor var_65_cast_fp16_8, tensor var_65_cast_fp16_9, tensor var_65_cast_fp16_10, tensor var_65_cast_fp16_11 = split(axis = var_65_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_65_cast_fp16")]; tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; tensor tile_1 = const()[name = string("tile_1"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351552)))]; int32 var_80_axis_0 = const()[name = string("op_80_axis_0"), val = int32(0)]; tensor var_80_cast_fp16_0, tensor var_80_cast_fp16_1, tensor var_80_cast_fp16_2, tensor var_80_cast_fp16_3, tensor var_80_cast_fp16_4, tensor var_80_cast_fp16_5, tensor var_80_cast_fp16_6, tensor var_80_cast_fp16_7, tensor var_80_cast_fp16_8, tensor var_80_cast_fp16_9, tensor var_80_cast_fp16_10, tensor var_80_cast_fp16_11 = split(axis = var_80_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_80_cast_fp16")]; tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 768, 1, 1536])]; tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 768, 1, 1536])]; tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; int32 var_108 = const()[name = string("op_108"), val = int32(3)]; tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; fp16 var_133_to_fp16 = const()[name = string("op_133_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_133_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351680)))]; tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80353280)))]; 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(80354880)))]; 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(80356480)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_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(80358080)))]; 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(81537792)))]; 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_5_cast_fp16)[name = string("query_1_cast_fp16")]; string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; int32 current_key_1_groups_0 = const()[name = string("current_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(81539392)))]; tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; int32 current_value_1_groups_0 = const()[name = string("current_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(82719104)))]; 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(83898816)))]; tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; tensor var_168_axes_0 = const()[name = string("op_168_axes_0"), val = tensor([1])]; tensor var_168_cast_fp16 = expand_dims(axes = var_168_axes_0, x = kv_cache_update_mask)[name = string("op_168_cast_fp16")]; tensor var_169_axes_0 = const()[name = string("op_169_axes_0"), val = tensor([2])]; tensor var_169_cast_fp16 = expand_dims(axes = var_169_axes_0, x = var_168_cast_fp16)[name = string("op_169_cast_fp16")]; tensor var_171_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_171_cast_fp16")]; tensor key_1_cast_fp16 = add(x = var_65_cast_fp16_0, y = var_171_cast_fp16)[name = string("key_1_cast_fp16")]; tensor var_173_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_173_cast_fp16")]; tensor value_1_cast_fp16 = add(x = var_80_cast_fp16_0, y = var_173_cast_fp16)[name = string("value_1_cast_fp16")]; tensor var_176 = const()[name = string("op_176"), val = tensor([1, 12, 64, -1])]; tensor mh_q_1_cast_fp16 = reshape(shape = var_176, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; fp16 var_178_to_fp16 = const()[name = string("op_178_to_fp16"), val = fp16(0x1p-3)]; tensor var_179_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_178_to_fp16)[name = string("op_179_cast_fp16")]; tensor var_180 = const()[name = string("op_180"), val = tensor([1, 12, 64, -1])]; tensor var_181_cast_fp16 = reshape(shape = var_180, x = key_1_cast_fp16)[name = string("op_181_cast_fp16")]; bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_179_cast_fp16, y = var_181_cast_fp16)[name = string("mh_w_1_cast_fp16")]; tensor var_185_axes_0 = const()[name = string("op_185_axes_0"), val = tensor([1])]; tensor var_185_cast_fp16 = expand_dims(axes = var_185_axes_0, x = decoder_key_padding_mask)[name = string("op_185_cast_fp16")]; tensor var_186_axes_0 = const()[name = string("op_186_axes_0"), val = tensor([2])]; tensor var_186_cast_fp16 = expand_dims(axes = var_186_axes_0, x = var_185_cast_fp16)[name = string("op_186_cast_fp16")]; tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_3_cast_fp16")]; tensor var_189_cast_fp16 = softmax(axis = var_108, x = mh_w_3_cast_fp16)[name = string("op_189_cast_fp16")]; tensor var_190 = const()[name = string("op_190"), val = tensor([1, 12, 64, -1])]; tensor var_191_cast_fp16 = reshape(shape = var_190, x = value_1_cast_fp16)[name = string("op_191_cast_fp16")]; bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_191_cast_fp16, y = var_189_cast_fp16)[name = string("attn_1_cast_fp16")]; tensor var_194 = const()[name = string("op_194"), val = tensor([1, 768, 1, -1])]; tensor input_1_cast_fp16 = reshape(shape = var_194, x = attn_1_cast_fp16)[name = string("input_1_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_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(83900416)))]; 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(85080128)))]; tensor obj_11_cast_fp16 = conv(bias = layers_0_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_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_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_216_to_fp16 = const()[name = string("op_216_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_216_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_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(85081728)))]; 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(85083328)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_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_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85084928)))]; tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86264640)))]; tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_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_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; tensor var_236 = const()[name = string("op_236"), val = tensor([1, 12, 64, -1])]; tensor mh_q_3_cast_fp16 = reshape(shape = var_236, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; fp16 var_238_to_fp16 = const()[name = string("op_238_to_fp16"), val = fp16(0x1p-3)]; tensor var_239_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_238_to_fp16)[name = string("op_239_cast_fp16")]; tensor var_240 = const()[name = string("op_240"), val = tensor([1, 12, 64, -1])]; tensor var_241_cast_fp16 = reshape(shape = var_240, x = obj_17_cast_fp16)[name = string("op_241_cast_fp16")]; bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_239_cast_fp16, y = var_241_cast_fp16)[name = string("mh_w_5_cast_fp16")]; tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; tensor var_245_axes_0 = const()[name = string("op_245_axes_0"), val = tensor([1])]; tensor var_245_cast_fp16 = expand_dims(axes = var_245_axes_0, x = read_state_4)[name = string("op_245_cast_fp16")]; tensor var_246_axes_0 = const()[name = string("op_246_axes_0"), val = tensor([2])]; tensor var_246_cast_fp16 = expand_dims(axes = var_246_axes_0, x = var_245_cast_fp16)[name = string("op_246_cast_fp16")]; tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_7_cast_fp16")]; tensor obj_23_cast_fp16 = softmax(axis = var_108, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; tensor var_250 = const()[name = string("op_250"), val = tensor([1, 12, 64, -1])]; tensor var_251_cast_fp16 = reshape(shape = var_250, x = obj_19_cast_fp16)[name = string("op_251_cast_fp16")]; bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_251_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; tensor var_254 = const()[name = string("op_254"), val = tensor([1, 768, 1, -1])]; tensor input_3_cast_fp16 = reshape(shape = var_254, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86266240)))]; tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87445952)))]; tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; fp16 var_272_to_fp16 = const()[name = string("op_272_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_272_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87447552)))]; tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87449152)))]; fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; int32 input_7_groups_0 = const()[name = string("input_7_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(87450752)))]; 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(92169408)))]; tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_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(92175616)))]; 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(96894272)))]; tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 768, 1, 1536])]; tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 768, 1, 1536])]; tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; int32 var_317 = const()[name = string("op_317"), val = int32(3)]; tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; fp16 var_342_to_fp16 = const()[name = string("op_342_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_342_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_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(96895872)))]; 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(96897472)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_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_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(96899072)))]; 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(98078784)))]; tensor query_5_cast_fp16 = conv(bias = layers_1_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_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; int32 current_key_3_groups_0 = const()[name = string("current_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(98080384)))]; tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; int32 current_value_3_groups_0 = const()[name = string("current_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(99260096)))]; 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(100439808)))]; tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; tensor var_380_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_380_cast_fp16")]; tensor key_3_cast_fp16 = add(x = var_65_cast_fp16_1, y = var_380_cast_fp16)[name = string("key_3_cast_fp16")]; tensor var_382_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_382_cast_fp16")]; tensor value_3_cast_fp16 = add(x = var_80_cast_fp16_1, y = var_382_cast_fp16)[name = string("value_3_cast_fp16")]; tensor var_385 = const()[name = string("op_385"), val = tensor([1, 12, 64, -1])]; tensor mh_q_5_cast_fp16 = reshape(shape = var_385, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; fp16 var_387_to_fp16 = const()[name = string("op_387_to_fp16"), val = fp16(0x1p-3)]; tensor var_388_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_387_to_fp16)[name = string("op_388_cast_fp16")]; tensor var_389 = const()[name = string("op_389"), val = tensor([1, 12, 64, -1])]; tensor var_390_cast_fp16 = reshape(shape = var_389, x = key_3_cast_fp16)[name = string("op_390_cast_fp16")]; bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_388_cast_fp16, y = var_390_cast_fp16)[name = string("mh_w_9_cast_fp16")]; tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_11_cast_fp16")]; tensor var_398_cast_fp16 = softmax(axis = var_317, x = mh_w_11_cast_fp16)[name = string("op_398_cast_fp16")]; tensor var_399 = const()[name = string("op_399"), val = tensor([1, 12, 64, -1])]; tensor var_400_cast_fp16 = reshape(shape = var_399, x = value_3_cast_fp16)[name = string("op_400_cast_fp16")]; bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_400_cast_fp16, y = var_398_cast_fp16)[name = string("attn_5_cast_fp16")]; tensor var_403 = const()[name = string("op_403"), val = tensor([1, 768, 1, -1])]; tensor input_11_cast_fp16 = reshape(shape = var_403, x = attn_5_cast_fp16)[name = string("input_11_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_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(100441408)))]; 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(101621120)))]; tensor obj_31_cast_fp16 = conv(bias = layers_1_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_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; fp16 var_425_to_fp16 = const()[name = string("op_425_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_425_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_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(101622720)))]; 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(101624320)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_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_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101625920)))]; tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102805632)))]; tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_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_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; tensor var_445 = const()[name = string("op_445"), val = tensor([1, 12, 64, -1])]; tensor mh_q_7_cast_fp16 = reshape(shape = var_445, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; fp16 var_447_to_fp16 = const()[name = string("op_447_to_fp16"), val = fp16(0x1p-3)]; tensor var_448_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_447_to_fp16)[name = string("op_448_cast_fp16")]; tensor var_449 = const()[name = string("op_449"), val = tensor([1, 12, 64, -1])]; tensor var_450_cast_fp16 = reshape(shape = var_449, x = obj_35_cast_fp16)[name = string("op_450_cast_fp16")]; bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_448_cast_fp16, y = var_450_cast_fp16)[name = string("mh_w_13_cast_fp16")]; tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_15_cast_fp16")]; tensor obj_41_cast_fp16 = softmax(axis = var_317, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; tensor var_459 = const()[name = string("op_459"), val = tensor([1, 12, 64, -1])]; tensor var_460_cast_fp16 = reshape(shape = var_459, x = obj_37_cast_fp16)[name = string("op_460_cast_fp16")]; bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_460_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; tensor var_463 = const()[name = string("op_463"), val = tensor([1, 768, 1, -1])]; tensor input_13_cast_fp16 = reshape(shape = var_463, x = attn_7_cast_fp16)[name = string("input_13_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_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102807232)))]; tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103986944)))]; tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_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_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_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_481_to_fp16 = const()[name = string("op_481_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_481_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103988544)))]; tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103990144)))]; fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; int32 input_17_groups_0 = const()[name = string("input_17_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(103991744)))]; 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(108710400)))]; tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_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_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(108716608)))]; 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(113435264)))]; tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_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_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 768, 1, 1536])]; tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 768, 1, 1536])]; tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; int32 var_526 = const()[name = string("op_526"), val = int32(3)]; tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; fp16 var_551_to_fp16 = const()[name = string("op_551_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_551_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113436864)))]; tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113438464)))]; fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_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_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(113440064)))]; 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(114619776)))]; tensor query_9_cast_fp16 = conv(bias = layers_2_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_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; int32 current_key_5_groups_0 = const()[name = string("current_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(114621376)))]; tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; int32 current_value_5_groups_0 = const()[name = string("current_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(115801088)))]; 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(116980800)))]; tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; tensor var_589_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_589_cast_fp16")]; tensor key_5_cast_fp16 = add(x = var_65_cast_fp16_2, y = var_589_cast_fp16)[name = string("key_5_cast_fp16")]; tensor var_591_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_591_cast_fp16")]; tensor value_5_cast_fp16 = add(x = var_80_cast_fp16_2, y = var_591_cast_fp16)[name = string("value_5_cast_fp16")]; tensor var_594 = const()[name = string("op_594"), val = tensor([1, 12, 64, -1])]; tensor mh_q_9_cast_fp16 = reshape(shape = var_594, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; tensor var_597_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_596_to_fp16)[name = string("op_597_cast_fp16")]; tensor var_598 = const()[name = string("op_598"), val = tensor([1, 12, 64, -1])]; tensor var_599_cast_fp16 = reshape(shape = var_598, x = key_5_cast_fp16)[name = string("op_599_cast_fp16")]; bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_597_cast_fp16, y = var_599_cast_fp16)[name = string("mh_w_17_cast_fp16")]; tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_19_cast_fp16")]; tensor var_607_cast_fp16 = softmax(axis = var_526, x = mh_w_19_cast_fp16)[name = string("op_607_cast_fp16")]; tensor var_608 = const()[name = string("op_608"), val = tensor([1, 12, 64, -1])]; tensor var_609_cast_fp16 = reshape(shape = var_608, x = value_5_cast_fp16)[name = string("op_609_cast_fp16")]; bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_609_cast_fp16, y = var_607_cast_fp16)[name = string("attn_9_cast_fp16")]; tensor var_612 = const()[name = string("op_612"), val = tensor([1, 768, 1, -1])]; tensor input_21_cast_fp16 = reshape(shape = var_612, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; int32 obj_49_groups_0 = const()[name = string("obj_49_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(116982400)))]; 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(118162112)))]; tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_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_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_634_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118163712)))]; tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118165312)))]; fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_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_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118166912)))]; tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119346624)))]; tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_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_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; tensor var_654 = const()[name = string("op_654"), val = tensor([1, 12, 64, -1])]; tensor mh_q_11_cast_fp16 = reshape(shape = var_654, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; fp16 var_656_to_fp16 = const()[name = string("op_656_to_fp16"), val = fp16(0x1p-3)]; tensor var_657_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_656_to_fp16)[name = string("op_657_cast_fp16")]; tensor var_658 = const()[name = string("op_658"), val = tensor([1, 12, 64, -1])]; tensor var_659_cast_fp16 = reshape(shape = var_658, x = obj_53_cast_fp16)[name = string("op_659_cast_fp16")]; bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_657_cast_fp16, y = var_659_cast_fp16)[name = string("mh_w_21_cast_fp16")]; tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_23_cast_fp16")]; tensor obj_59_cast_fp16 = softmax(axis = var_526, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; tensor var_668 = const()[name = string("op_668"), val = tensor([1, 12, 64, -1])]; tensor var_669_cast_fp16 = reshape(shape = var_668, x = obj_55_cast_fp16)[name = string("op_669_cast_fp16")]; bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_669_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; tensor var_672 = const()[name = string("op_672"), val = tensor([1, 768, 1, -1])]; tensor input_23_cast_fp16 = reshape(shape = var_672, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119348224)))]; tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120527936)))]; tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; fp16 var_690_to_fp16 = const()[name = string("op_690_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_690_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120529536)))]; tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120531136)))]; fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; int32 input_27_groups_0 = const()[name = string("input_27_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(120532736)))]; 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(125251392)))]; tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_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_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(125257600)))]; 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(129976256)))]; tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_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_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 768, 1, 1536])]; tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 768, 1, 1536])]; tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; int32 var_735 = const()[name = string("op_735"), val = int32(3)]; tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; fp16 var_760_to_fp16 = const()[name = string("op_760_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_760_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129977856)))]; tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129979456)))]; fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_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_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(129981056)))]; 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(131160768)))]; tensor query_13_cast_fp16 = conv(bias = layers_3_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_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")]; tensor current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor([1, 1])]; tensor current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor([1, 1])]; int32 current_key_7_groups_0 = const()[name = string("current_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(131162368)))]; tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_7_cast_fp16")]; string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")]; tensor current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor([1, 1])]; tensor current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor([1, 1])]; int32 current_value_7_groups_0 = const()[name = string("current_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(132342080)))]; 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(133521792)))]; tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_7_cast_fp16")]; tensor var_798_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_798_cast_fp16")]; tensor key_7_cast_fp16 = add(x = var_65_cast_fp16_3, y = var_798_cast_fp16)[name = string("key_7_cast_fp16")]; tensor var_800_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_800_cast_fp16")]; tensor value_7_cast_fp16 = add(x = var_80_cast_fp16_3, y = var_800_cast_fp16)[name = string("value_7_cast_fp16")]; tensor var_803 = const()[name = string("op_803"), val = tensor([1, 12, 64, -1])]; tensor mh_q_13_cast_fp16 = reshape(shape = var_803, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; fp16 var_805_to_fp16 = const()[name = string("op_805_to_fp16"), val = fp16(0x1p-3)]; tensor var_806_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_805_to_fp16)[name = string("op_806_cast_fp16")]; tensor var_807 = const()[name = string("op_807"), val = tensor([1, 12, 64, -1])]; tensor var_808_cast_fp16 = reshape(shape = var_807, x = key_7_cast_fp16)[name = string("op_808_cast_fp16")]; bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_806_cast_fp16, y = var_808_cast_fp16)[name = string("mh_w_25_cast_fp16")]; tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_27_cast_fp16")]; tensor var_816_cast_fp16 = softmax(axis = var_735, x = mh_w_27_cast_fp16)[name = string("op_816_cast_fp16")]; tensor var_817 = const()[name = string("op_817"), val = tensor([1, 12, 64, -1])]; tensor var_818_cast_fp16 = reshape(shape = var_817, x = value_7_cast_fp16)[name = string("op_818_cast_fp16")]; bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_818_cast_fp16, y = var_816_cast_fp16)[name = string("attn_13_cast_fp16")]; tensor var_821 = const()[name = string("op_821"), val = tensor([1, 768, 1, -1])]; tensor input_31_cast_fp16 = reshape(shape = var_821, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; int32 obj_67_groups_0 = const()[name = string("obj_67_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(133523392)))]; 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(134703104)))]; tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; fp16 var_843_to_fp16 = const()[name = string("op_843_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_843_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134704704)))]; tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134706304)))]; fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_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_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134707904)))]; tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135887616)))]; tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_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_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_15_cast_fp16")]; tensor var_863 = const()[name = string("op_863"), val = tensor([1, 12, 64, -1])]; tensor mh_q_15_cast_fp16 = reshape(shape = var_863, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; fp16 var_865_to_fp16 = const()[name = string("op_865_to_fp16"), val = fp16(0x1p-3)]; tensor var_866_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_865_to_fp16)[name = string("op_866_cast_fp16")]; tensor var_867 = const()[name = string("op_867"), val = tensor([1, 12, 64, -1])]; tensor var_868_cast_fp16 = reshape(shape = var_867, x = obj_71_cast_fp16)[name = string("op_868_cast_fp16")]; bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_866_cast_fp16, y = var_868_cast_fp16)[name = string("mh_w_29_cast_fp16")]; tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_31_cast_fp16")]; tensor obj_77_cast_fp16 = softmax(axis = var_735, x = mh_w_31_cast_fp16)[name = string("obj_77_cast_fp16")]; tensor var_877 = const()[name = string("op_877"), val = tensor([1, 12, 64, -1])]; tensor var_878_cast_fp16 = reshape(shape = var_877, x = obj_73_cast_fp16)[name = string("op_878_cast_fp16")]; bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_878_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_15_cast_fp16")]; tensor var_881 = const()[name = string("op_881"), val = tensor([1, 768, 1, -1])]; tensor input_33_cast_fp16 = reshape(shape = var_881, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")]; string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135889216)))]; tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137068928)))]; tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_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_899_to_fp16 = const()[name = string("op_899_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_899_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_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(137070528)))]; 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(137072128)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_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_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(137073728)))]; 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(141792384)))]; tensor input_37_cast_fp16 = conv(bias = layers_3_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_3_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_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_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(141798592)))]; 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(146517248)))]; tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_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_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")]; tensor obj_89_begin_0 = const()[name = string("obj_89_begin_0"), val = tensor([4, 0, 0, 0])]; tensor obj_89_end_0 = const()[name = string("obj_89_end_0"), val = tensor([5, 768, 1, 1536])]; tensor obj_89_end_mask_0 = const()[name = string("obj_89_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_89_cast_fp16 = slice_by_index(begin = obj_89_begin_0, end = obj_89_end_0, end_mask = obj_89_end_mask_0, x = read_state_2)[name = string("obj_89_cast_fp16")]; tensor obj_91_begin_0 = const()[name = string("obj_91_begin_0"), val = tensor([4, 0, 0, 0])]; tensor obj_91_end_0 = const()[name = string("obj_91_end_0"), val = tensor([5, 768, 1, 1536])]; tensor obj_91_end_mask_0 = const()[name = string("obj_91_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_91_cast_fp16 = slice_by_index(begin = obj_91_begin_0, end = obj_91_end_0, end_mask = obj_91_end_mask_0, x = read_state_3)[name = string("obj_91_cast_fp16")]; int32 var_944 = const()[name = string("op_944"), val = int32(3)]; tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; fp16 var_969_to_fp16 = const()[name = string("op_969_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_969_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146518848)))]; tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146520448)))]; fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_79_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_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(146522048)))]; 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(147701760)))]; tensor query_17_cast_fp16 = conv(bias = layers_4_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_4_self_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("query_17_cast_fp16")]; string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")]; tensor current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor([1, 1])]; tensor current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor([1, 1])]; int32 current_key_9_groups_0 = const()[name = string("current_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(147703360)))]; tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_key_9_cast_fp16")]; string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")]; tensor current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor([1, 1])]; tensor current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor([1, 1])]; int32 current_value_9_groups_0 = const()[name = string("current_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(148883072)))]; 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(150062784)))]; tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_value_9_cast_fp16")]; tensor var_1007_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1007_cast_fp16")]; tensor key_9_cast_fp16 = add(x = var_65_cast_fp16_4, y = var_1007_cast_fp16)[name = string("key_9_cast_fp16")]; tensor var_1009_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1009_cast_fp16")]; tensor value_9_cast_fp16 = add(x = var_80_cast_fp16_4, y = var_1009_cast_fp16)[name = string("value_9_cast_fp16")]; tensor var_1012 = const()[name = string("op_1012"), val = tensor([1, 12, 64, -1])]; tensor mh_q_17_cast_fp16 = reshape(shape = var_1012, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; fp16 var_1014_to_fp16 = const()[name = string("op_1014_to_fp16"), val = fp16(0x1p-3)]; tensor var_1015_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1014_to_fp16)[name = string("op_1015_cast_fp16")]; tensor var_1016 = const()[name = string("op_1016"), val = tensor([1, 12, 64, -1])]; tensor var_1017_cast_fp16 = reshape(shape = var_1016, x = key_9_cast_fp16)[name = string("op_1017_cast_fp16")]; bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_1015_cast_fp16, y = var_1017_cast_fp16)[name = string("mh_w_33_cast_fp16")]; tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_35_cast_fp16")]; tensor var_1025_cast_fp16 = softmax(axis = var_944, x = mh_w_35_cast_fp16)[name = string("op_1025_cast_fp16")]; tensor var_1026 = const()[name = string("op_1026"), val = tensor([1, 12, 64, -1])]; tensor var_1027_cast_fp16 = reshape(shape = var_1026, x = value_9_cast_fp16)[name = string("op_1027_cast_fp16")]; bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1027_cast_fp16, y = var_1025_cast_fp16)[name = string("attn_17_cast_fp16")]; tensor var_1030 = const()[name = string("op_1030"), val = tensor([1, 768, 1, -1])]; tensor input_41_cast_fp16 = reshape(shape = var_1030, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")]; string obj_85_pad_type_0 = const()[name = string("obj_85_pad_type_0"), val = string("valid")]; tensor obj_85_strides_0 = const()[name = string("obj_85_strides_0"), val = tensor([1, 1])]; tensor obj_85_pad_0 = const()[name = string("obj_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_85_dilations_0 = const()[name = string("obj_85_dilations_0"), val = tensor([1, 1])]; int32 obj_85_groups_0 = const()[name = string("obj_85_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(150064384)))]; 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(151244096)))]; tensor obj_85_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_85_dilations_0, groups = obj_85_groups_0, pad = obj_85_pad_0, pad_type = obj_85_pad_type_0, strides = obj_85_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_85_cast_fp16")]; tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_85_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_1052_to_fp16 = const()[name = string("op_1052_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1052_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151245696)))]; tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151247296)))]; fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_87_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_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151248896)))]; tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152428608)))]; tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_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_4_encoder_attn_q_proj_weight_to_fp16, x = obj_87_cast_fp16)[name = string("query_19_cast_fp16")]; tensor var_1072 = const()[name = string("op_1072"), val = tensor([1, 12, 64, -1])]; tensor mh_q_19_cast_fp16 = reshape(shape = var_1072, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; fp16 var_1074_to_fp16 = const()[name = string("op_1074_to_fp16"), val = fp16(0x1p-3)]; tensor var_1075_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1074_to_fp16)[name = string("op_1075_cast_fp16")]; tensor var_1076 = const()[name = string("op_1076"), val = tensor([1, 12, 64, -1])]; tensor var_1077_cast_fp16 = reshape(shape = var_1076, x = obj_89_cast_fp16)[name = string("op_1077_cast_fp16")]; bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1075_cast_fp16, y = var_1077_cast_fp16)[name = string("mh_w_37_cast_fp16")]; tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_39_cast_fp16")]; tensor obj_95_cast_fp16 = softmax(axis = var_944, x = mh_w_39_cast_fp16)[name = string("obj_95_cast_fp16")]; tensor var_1086 = const()[name = string("op_1086"), val = tensor([1, 12, 64, -1])]; tensor var_1087_cast_fp16 = reshape(shape = var_1086, x = obj_91_cast_fp16)[name = string("op_1087_cast_fp16")]; bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1087_cast_fp16, y = obj_95_cast_fp16)[name = string("attn_19_cast_fp16")]; tensor var_1090 = const()[name = string("op_1090"), val = tensor([1, 768, 1, -1])]; tensor input_43_cast_fp16 = reshape(shape = var_1090, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")]; string obj_93_pad_type_0 = const()[name = string("obj_93_pad_type_0"), val = string("valid")]; tensor obj_93_strides_0 = const()[name = string("obj_93_strides_0"), val = tensor([1, 1])]; tensor obj_93_pad_0 = const()[name = string("obj_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_93_dilations_0 = const()[name = string("obj_93_dilations_0"), val = tensor([1, 1])]; int32 obj_93_groups_0 = const()[name = string("obj_93_groups_0"), val = int32(1)]; tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152430208)))]; tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153609920)))]; tensor obj_93_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_93_dilations_0, groups = obj_93_groups_0, pad = obj_93_pad_0, pad_type = obj_93_pad_type_0, strides = obj_93_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("obj_93_cast_fp16")]; tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_29_cast_fp16")]; tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; fp16 var_1108_to_fp16 = const()[name = string("op_1108_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1108_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; tensor input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611520)))]; tensor input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153613120)))]; fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")]; string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; int32 input_47_groups_0 = const()[name = string("input_47_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(153614720)))]; 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(158333376)))]; tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_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_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(158339584)))]; 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(163058240)))]; tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_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_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")]; tensor obj_107_begin_0 = const()[name = string("obj_107_begin_0"), val = tensor([5, 0, 0, 0])]; tensor obj_107_end_0 = const()[name = string("obj_107_end_0"), val = tensor([6, 768, 1, 1536])]; tensor obj_107_end_mask_0 = const()[name = string("obj_107_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_107_cast_fp16 = slice_by_index(begin = obj_107_begin_0, end = obj_107_end_0, end_mask = obj_107_end_mask_0, x = read_state_2)[name = string("obj_107_cast_fp16")]; tensor obj_109_begin_0 = const()[name = string("obj_109_begin_0"), val = tensor([5, 0, 0, 0])]; tensor obj_109_end_0 = const()[name = string("obj_109_end_0"), val = tensor([6, 768, 1, 1536])]; tensor obj_109_end_mask_0 = const()[name = string("obj_109_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_109_cast_fp16 = slice_by_index(begin = obj_109_begin_0, end = obj_109_end_0, end_mask = obj_109_end_mask_0, x = read_state_3)[name = string("obj_109_cast_fp16")]; int32 var_1153 = const()[name = string("op_1153"), val = int32(3)]; tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; fp16 var_1178_to_fp16 = const()[name = string("op_1178_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1178_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163059840)))]; tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163061440)))]; fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_97_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_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(163063040)))]; 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(164242752)))]; tensor query_21_cast_fp16 = conv(bias = layers_5_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_5_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_21_cast_fp16")]; string current_key_11_pad_type_0 = const()[name = string("current_key_11_pad_type_0"), val = string("valid")]; tensor current_key_11_strides_0 = const()[name = string("current_key_11_strides_0"), val = tensor([1, 1])]; tensor current_key_11_pad_0 = const()[name = string("current_key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_11_dilations_0 = const()[name = string("current_key_11_dilations_0"), val = tensor([1, 1])]; int32 current_key_11_groups_0 = const()[name = string("current_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(164244352)))]; tensor current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_key_11_cast_fp16")]; string current_value_11_pad_type_0 = const()[name = string("current_value_11_pad_type_0"), val = string("valid")]; tensor current_value_11_strides_0 = const()[name = string("current_value_11_strides_0"), val = tensor([1, 1])]; tensor current_value_11_pad_0 = const()[name = string("current_value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_11_dilations_0 = const()[name = string("current_value_11_dilations_0"), val = tensor([1, 1])]; int32 current_value_11_groups_0 = const()[name = string("current_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(165424064)))]; 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(166603776)))]; tensor current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_value_11_cast_fp16")]; tensor var_1216_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1216_cast_fp16")]; tensor key_11_cast_fp16 = add(x = var_65_cast_fp16_5, y = var_1216_cast_fp16)[name = string("key_11_cast_fp16")]; tensor var_1218_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1218_cast_fp16")]; tensor value_11_cast_fp16 = add(x = var_80_cast_fp16_5, y = var_1218_cast_fp16)[name = string("value_11_cast_fp16")]; tensor var_1221 = const()[name = string("op_1221"), val = tensor([1, 12, 64, -1])]; tensor mh_q_21_cast_fp16 = reshape(shape = var_1221, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; fp16 var_1223_to_fp16 = const()[name = string("op_1223_to_fp16"), val = fp16(0x1p-3)]; tensor var_1224_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1223_to_fp16)[name = string("op_1224_cast_fp16")]; tensor var_1225 = const()[name = string("op_1225"), val = tensor([1, 12, 64, -1])]; tensor var_1226_cast_fp16 = reshape(shape = var_1225, x = key_11_cast_fp16)[name = string("op_1226_cast_fp16")]; bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1224_cast_fp16, y = var_1226_cast_fp16)[name = string("mh_w_41_cast_fp16")]; tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_43_cast_fp16")]; tensor var_1234_cast_fp16 = softmax(axis = var_1153, x = mh_w_43_cast_fp16)[name = string("op_1234_cast_fp16")]; tensor var_1235 = const()[name = string("op_1235"), val = tensor([1, 12, 64, -1])]; tensor var_1236_cast_fp16 = reshape(shape = var_1235, x = value_11_cast_fp16)[name = string("op_1236_cast_fp16")]; bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1236_cast_fp16, y = var_1234_cast_fp16)[name = string("attn_21_cast_fp16")]; tensor var_1239 = const()[name = string("op_1239"), val = tensor([1, 768, 1, -1])]; tensor input_51_cast_fp16 = reshape(shape = var_1239, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")]; string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; int32 obj_103_groups_0 = const()[name = string("obj_103_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(166605376)))]; 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(167785088)))]; tensor obj_103_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("obj_103_cast_fp16")]; tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_33_cast_fp16")]; tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1261_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167786688)))]; tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167788288)))]; fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_105_cast_fp16")]; string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167789888)))]; tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168969600)))]; tensor query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_23_cast_fp16")]; tensor var_1281 = const()[name = string("op_1281"), val = tensor([1, 12, 64, -1])]; tensor mh_q_23_cast_fp16 = reshape(shape = var_1281, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")]; fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; tensor var_1284_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1283_to_fp16)[name = string("op_1284_cast_fp16")]; tensor var_1285 = const()[name = string("op_1285"), val = tensor([1, 12, 64, -1])]; tensor var_1286_cast_fp16 = reshape(shape = var_1285, x = obj_107_cast_fp16)[name = string("op_1286_cast_fp16")]; bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_1284_cast_fp16, y = var_1286_cast_fp16)[name = string("mh_w_45_cast_fp16")]; tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_47_cast_fp16")]; tensor obj_113_cast_fp16 = softmax(axis = var_1153, x = mh_w_47_cast_fp16)[name = string("obj_113_cast_fp16")]; tensor var_1295 = const()[name = string("op_1295"), val = tensor([1, 12, 64, -1])]; tensor var_1296_cast_fp16 = reshape(shape = var_1295, x = obj_109_cast_fp16)[name = string("op_1296_cast_fp16")]; bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1296_cast_fp16, y = obj_113_cast_fp16)[name = string("attn_23_cast_fp16")]; tensor var_1299 = const()[name = string("op_1299"), val = tensor([1, 768, 1, -1])]; tensor input_53_cast_fp16 = reshape(shape = var_1299, x = attn_23_cast_fp16)[name = string("input_53_cast_fp16")]; string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168971200)))]; tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170150912)))]; tensor obj_111_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("obj_111_cast_fp16")]; tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_111_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_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1317_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; tensor input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170152512)))]; tensor input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170154112)))]; fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")]; string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; int32 input_57_groups_0 = const()[name = string("input_57_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(170155712)))]; 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(174874368)))]; tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; string input_59_mode_0 = const()[name = string("input_59_mode_0"), val = string("EXACT")]; tensor input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = string("input_59_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_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(174880576)))]; 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(179599232)))]; tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_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_5_fc2_weight_to_fp16, x = input_59_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_37_cast_fp16")]; tensor obj_125_begin_0 = const()[name = string("obj_125_begin_0"), val = tensor([6, 0, 0, 0])]; tensor obj_125_end_0 = const()[name = string("obj_125_end_0"), val = tensor([7, 768, 1, 1536])]; tensor obj_125_end_mask_0 = const()[name = string("obj_125_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_125_cast_fp16 = slice_by_index(begin = obj_125_begin_0, end = obj_125_end_0, end_mask = obj_125_end_mask_0, x = read_state_2)[name = string("obj_125_cast_fp16")]; tensor obj_127_begin_0 = const()[name = string("obj_127_begin_0"), val = tensor([6, 0, 0, 0])]; tensor obj_127_end_0 = const()[name = string("obj_127_end_0"), val = tensor([7, 768, 1, 1536])]; tensor obj_127_end_mask_0 = const()[name = string("obj_127_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_127_cast_fp16 = slice_by_index(begin = obj_127_begin_0, end = obj_127_end_0, end_mask = obj_127_end_mask_0, x = read_state_3)[name = string("obj_127_cast_fp16")]; int32 var_1362 = const()[name = string("op_1362"), val = int32(3)]; tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; fp16 var_1387_to_fp16 = const()[name = string("op_1387_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1387_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; tensor obj_115_gamma_0_to_fp16 = const()[name = string("obj_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179600832)))]; tensor obj_115_beta_0_to_fp16 = const()[name = string("obj_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179602432)))]; fp16 obj_115_epsilon_0_to_fp16 = const()[name = string("obj_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_115_cast_fp16 = batch_norm(beta = obj_115_beta_0_to_fp16, epsilon = obj_115_epsilon_0_to_fp16, gamma = obj_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_115_cast_fp16")]; string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; int32 query_25_groups_0 = const()[name = string("query_25_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(179604032)))]; 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(180783744)))]; tensor query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("query_25_cast_fp16")]; string current_key_13_pad_type_0 = const()[name = string("current_key_13_pad_type_0"), val = string("valid")]; tensor current_key_13_strides_0 = const()[name = string("current_key_13_strides_0"), val = tensor([1, 1])]; tensor current_key_13_pad_0 = const()[name = string("current_key_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_13_dilations_0 = const()[name = string("current_key_13_dilations_0"), val = tensor([1, 1])]; int32 current_key_13_groups_0 = const()[name = string("current_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(180785344)))]; tensor current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_key_13_cast_fp16")]; string current_value_13_pad_type_0 = const()[name = string("current_value_13_pad_type_0"), val = string("valid")]; tensor current_value_13_strides_0 = const()[name = string("current_value_13_strides_0"), val = tensor([1, 1])]; tensor current_value_13_pad_0 = const()[name = string("current_value_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_13_dilations_0 = const()[name = string("current_value_13_dilations_0"), val = tensor([1, 1])]; int32 current_value_13_groups_0 = const()[name = string("current_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(181965056)))]; 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(183144768)))]; tensor current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_value_13_cast_fp16")]; tensor var_1425_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1425_cast_fp16")]; tensor key_13_cast_fp16 = add(x = var_65_cast_fp16_6, y = var_1425_cast_fp16)[name = string("key_13_cast_fp16")]; tensor var_1427_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1427_cast_fp16")]; tensor value_13_cast_fp16 = add(x = var_80_cast_fp16_6, y = var_1427_cast_fp16)[name = string("value_13_cast_fp16")]; tensor var_1430 = const()[name = string("op_1430"), val = tensor([1, 12, 64, -1])]; tensor mh_q_25_cast_fp16 = reshape(shape = var_1430, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")]; fp16 var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = fp16(0x1p-3)]; tensor var_1433_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1432_to_fp16)[name = string("op_1433_cast_fp16")]; tensor var_1434 = const()[name = string("op_1434"), val = tensor([1, 12, 64, -1])]; tensor var_1435_cast_fp16 = reshape(shape = var_1434, x = key_13_cast_fp16)[name = string("op_1435_cast_fp16")]; bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1433_cast_fp16, y = var_1435_cast_fp16)[name = string("mh_w_49_cast_fp16")]; tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_51_cast_fp16")]; tensor var_1443_cast_fp16 = softmax(axis = var_1362, x = mh_w_51_cast_fp16)[name = string("op_1443_cast_fp16")]; tensor var_1444 = const()[name = string("op_1444"), val = tensor([1, 12, 64, -1])]; tensor var_1445_cast_fp16 = reshape(shape = var_1444, x = value_13_cast_fp16)[name = string("op_1445_cast_fp16")]; bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1445_cast_fp16, y = var_1443_cast_fp16)[name = string("attn_25_cast_fp16")]; tensor var_1448 = const()[name = string("op_1448"), val = tensor([1, 768, 1, -1])]; tensor input_61_cast_fp16 = reshape(shape = var_1448, x = attn_25_cast_fp16)[name = string("input_61_cast_fp16")]; string obj_121_pad_type_0 = const()[name = string("obj_121_pad_type_0"), val = string("valid")]; tensor obj_121_strides_0 = const()[name = string("obj_121_strides_0"), val = tensor([1, 1])]; tensor obj_121_pad_0 = const()[name = string("obj_121_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_121_dilations_0 = const()[name = string("obj_121_dilations_0"), val = tensor([1, 1])]; int32 obj_121_groups_0 = const()[name = string("obj_121_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(183146368)))]; 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(184326080)))]; tensor obj_121_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_121_dilations_0, groups = obj_121_groups_0, pad = obj_121_pad_0, pad_type = obj_121_pad_type_0, strides = obj_121_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_61_cast_fp16)[name = string("obj_121_cast_fp16")]; tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_121_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_1470_to_fp16 = const()[name = string("op_1470_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1470_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; tensor obj_123_gamma_0_to_fp16 = const()[name = string("obj_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184327680)))]; tensor obj_123_beta_0_to_fp16 = const()[name = string("obj_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184329280)))]; fp16 obj_123_epsilon_0_to_fp16 = const()[name = string("obj_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_123_cast_fp16 = batch_norm(beta = obj_123_beta_0_to_fp16, epsilon = obj_123_epsilon_0_to_fp16, gamma = obj_123_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("obj_123_cast_fp16")]; string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; tensor layers_6_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184330880)))]; tensor layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185510592)))]; tensor query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16, x = obj_123_cast_fp16)[name = string("query_27_cast_fp16")]; tensor var_1490 = const()[name = string("op_1490"), val = tensor([1, 12, 64, -1])]; tensor mh_q_27_cast_fp16 = reshape(shape = var_1490, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")]; fp16 var_1492_to_fp16 = const()[name = string("op_1492_to_fp16"), val = fp16(0x1p-3)]; tensor var_1493_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1492_to_fp16)[name = string("op_1493_cast_fp16")]; tensor var_1494 = const()[name = string("op_1494"), val = tensor([1, 12, 64, -1])]; tensor var_1495_cast_fp16 = reshape(shape = var_1494, x = obj_125_cast_fp16)[name = string("op_1495_cast_fp16")]; bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_1493_cast_fp16, y = var_1495_cast_fp16)[name = string("mh_w_53_cast_fp16")]; tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_55_cast_fp16")]; tensor obj_131_cast_fp16 = softmax(axis = var_1362, x = mh_w_55_cast_fp16)[name = string("obj_131_cast_fp16")]; tensor var_1504 = const()[name = string("op_1504"), val = tensor([1, 12, 64, -1])]; tensor var_1505_cast_fp16 = reshape(shape = var_1504, x = obj_127_cast_fp16)[name = string("op_1505_cast_fp16")]; bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1505_cast_fp16, y = obj_131_cast_fp16)[name = string("attn_27_cast_fp16")]; tensor var_1508 = const()[name = string("op_1508"), val = tensor([1, 768, 1, -1])]; tensor input_63_cast_fp16 = reshape(shape = var_1508, x = attn_27_cast_fp16)[name = string("input_63_cast_fp16")]; string obj_129_pad_type_0 = const()[name = string("obj_129_pad_type_0"), val = string("valid")]; tensor obj_129_strides_0 = const()[name = string("obj_129_strides_0"), val = tensor([1, 1])]; tensor obj_129_pad_0 = const()[name = string("obj_129_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_129_dilations_0 = const()[name = string("obj_129_dilations_0"), val = tensor([1, 1])]; int32 obj_129_groups_0 = const()[name = string("obj_129_groups_0"), val = int32(1)]; tensor layers_6_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185512192)))]; tensor layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186691904)))]; tensor obj_129_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_129_dilations_0, groups = obj_129_groups_0, pad = obj_129_pad_0, pad_type = obj_129_pad_type_0, strides = obj_129_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("obj_129_cast_fp16")]; tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_129_cast_fp16)[name = string("inputs_41_cast_fp16")]; tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; fp16 var_1529_to_fp16 = const()[name = string("op_1529_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1529_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; tensor input_65_gamma_0_to_fp16 = const()[name = string("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186693504)))]; tensor input_65_beta_0_to_fp16 = const()[name = string("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186695104)))]; fp16 input_65_epsilon_0_to_fp16 = const()[name = string("input_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("input_65_cast_fp16")]; string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")]; tensor input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor([1, 1])]; tensor input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor([1, 1])]; int32 input_67_groups_0 = const()[name = string("input_67_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(186696704)))]; 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(191415360)))]; tensor input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; string input_69_mode_0 = const()[name = string("input_69_mode_0"), val = string("EXACT")]; tensor input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = string("input_69_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_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(191421568)))]; 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(196140224)))]; tensor hidden_states_15_cast_fp16 = conv(bias = layers_6_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_6_fc2_weight_to_fp16, x = input_69_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_43_cast_fp16")]; tensor obj_143_begin_0 = const()[name = string("obj_143_begin_0"), val = tensor([7, 0, 0, 0])]; tensor obj_143_end_0 = const()[name = string("obj_143_end_0"), val = tensor([8, 768, 1, 1536])]; tensor obj_143_end_mask_0 = const()[name = string("obj_143_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_143_cast_fp16 = slice_by_index(begin = obj_143_begin_0, end = obj_143_end_0, end_mask = obj_143_end_mask_0, x = read_state_2)[name = string("obj_143_cast_fp16")]; tensor obj_145_begin_0 = const()[name = string("obj_145_begin_0"), val = tensor([7, 0, 0, 0])]; tensor obj_145_end_0 = const()[name = string("obj_145_end_0"), val = tensor([8, 768, 1, 1536])]; tensor obj_145_end_mask_0 = const()[name = string("obj_145_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_145_cast_fp16 = slice_by_index(begin = obj_145_begin_0, end = obj_145_end_0, end_mask = obj_145_end_mask_0, x = read_state_3)[name = string("obj_145_cast_fp16")]; int32 var_1575 = const()[name = string("op_1575"), val = int32(3)]; tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1600_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; tensor obj_133_gamma_0_to_fp16 = const()[name = string("obj_133_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196141824)))]; tensor obj_133_beta_0_to_fp16 = const()[name = string("obj_133_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196143424)))]; fp16 obj_133_epsilon_0_to_fp16 = const()[name = string("obj_133_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_133_cast_fp16 = batch_norm(beta = obj_133_beta_0_to_fp16, epsilon = obj_133_epsilon_0_to_fp16, gamma = obj_133_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("obj_133_cast_fp16")]; string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; int32 query_29_groups_0 = const()[name = string("query_29_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(196145024)))]; 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(197324736)))]; tensor query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("query_29_cast_fp16")]; string current_key_15_pad_type_0 = const()[name = string("current_key_15_pad_type_0"), val = string("valid")]; tensor current_key_15_strides_0 = const()[name = string("current_key_15_strides_0"), val = tensor([1, 1])]; tensor current_key_15_pad_0 = const()[name = string("current_key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_15_dilations_0 = const()[name = string("current_key_15_dilations_0"), val = tensor([1, 1])]; int32 current_key_15_groups_0 = const()[name = string("current_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(197326336)))]; tensor current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_key_15_cast_fp16")]; string current_value_15_pad_type_0 = const()[name = string("current_value_15_pad_type_0"), val = string("valid")]; tensor current_value_15_strides_0 = const()[name = string("current_value_15_strides_0"), val = tensor([1, 1])]; tensor current_value_15_pad_0 = const()[name = string("current_value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_15_dilations_0 = const()[name = string("current_value_15_dilations_0"), val = tensor([1, 1])]; int32 current_value_15_groups_0 = const()[name = string("current_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(198506048)))]; 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(199685760)))]; tensor current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_value_15_cast_fp16")]; tensor var_1638_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1638_cast_fp16")]; tensor key_15_cast_fp16 = add(x = var_65_cast_fp16_7, y = var_1638_cast_fp16)[name = string("key_15_cast_fp16")]; tensor var_1640_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1640_cast_fp16")]; tensor value_15_cast_fp16 = add(x = var_80_cast_fp16_7, y = var_1640_cast_fp16)[name = string("value_15_cast_fp16")]; tensor var_1643 = const()[name = string("op_1643"), val = tensor([1, 12, 64, -1])]; tensor mh_q_29_cast_fp16 = reshape(shape = var_1643, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")]; fp16 var_1645_to_fp16 = const()[name = string("op_1645_to_fp16"), val = fp16(0x1p-3)]; tensor var_1646_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1645_to_fp16)[name = string("op_1646_cast_fp16")]; tensor var_1647 = const()[name = string("op_1647"), val = tensor([1, 12, 64, -1])]; tensor var_1648_cast_fp16 = reshape(shape = var_1647, x = key_15_cast_fp16)[name = string("op_1648_cast_fp16")]; bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_1646_cast_fp16, y = var_1648_cast_fp16)[name = string("mh_w_57_cast_fp16")]; tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_59_cast_fp16")]; tensor var_1656_cast_fp16 = softmax(axis = var_1575, x = mh_w_59_cast_fp16)[name = string("op_1656_cast_fp16")]; tensor var_1657 = const()[name = string("op_1657"), val = tensor([1, 12, 64, -1])]; tensor var_1658_cast_fp16 = reshape(shape = var_1657, x = value_15_cast_fp16)[name = string("op_1658_cast_fp16")]; bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1658_cast_fp16, y = var_1656_cast_fp16)[name = string("attn_29_cast_fp16")]; tensor var_1661 = const()[name = string("op_1661"), val = tensor([1, 768, 1, -1])]; tensor input_71_cast_fp16 = reshape(shape = var_1661, x = attn_29_cast_fp16)[name = string("input_71_cast_fp16")]; string obj_139_pad_type_0 = const()[name = string("obj_139_pad_type_0"), val = string("valid")]; tensor obj_139_strides_0 = const()[name = string("obj_139_strides_0"), val = tensor([1, 1])]; tensor obj_139_pad_0 = const()[name = string("obj_139_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_139_dilations_0 = const()[name = string("obj_139_dilations_0"), val = tensor([1, 1])]; int32 obj_139_groups_0 = const()[name = string("obj_139_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(199687360)))]; 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(200867072)))]; tensor obj_139_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_139_dilations_0, groups = obj_139_groups_0, pad = obj_139_pad_0, pad_type = obj_139_pad_type_0, strides = obj_139_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("obj_139_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_139_cast_fp16)[name = string("inputs_45_cast_fp16")]; tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; fp16 var_1683_to_fp16 = const()[name = string("op_1683_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1683_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; tensor obj_141_gamma_0_to_fp16 = const()[name = string("obj_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200868672)))]; tensor obj_141_beta_0_to_fp16 = const()[name = string("obj_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200870272)))]; fp16 obj_141_epsilon_0_to_fp16 = const()[name = string("obj_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_141_cast_fp16")]; string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; tensor layers_7_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200871872)))]; tensor layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202051584)))]; tensor query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = string("query_31_cast_fp16")]; tensor var_1703 = const()[name = string("op_1703"), val = tensor([1, 12, 64, -1])]; tensor mh_q_31_cast_fp16 = reshape(shape = var_1703, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")]; fp16 var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = fp16(0x1p-3)]; tensor var_1706_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1705_to_fp16)[name = string("op_1706_cast_fp16")]; tensor var_1707 = const()[name = string("op_1707"), val = tensor([1, 12, 64, -1])]; tensor var_1708_cast_fp16 = reshape(shape = var_1707, x = obj_143_cast_fp16)[name = string("op_1708_cast_fp16")]; bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_1706_cast_fp16, y = var_1708_cast_fp16)[name = string("mh_w_61_cast_fp16")]; tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_63_cast_fp16")]; tensor obj_149_cast_fp16 = softmax(axis = var_1575, x = mh_w_63_cast_fp16)[name = string("obj_149_cast_fp16")]; tensor var_1717 = const()[name = string("op_1717"), val = tensor([1, 12, 64, -1])]; tensor var_1718_cast_fp16 = reshape(shape = var_1717, x = obj_145_cast_fp16)[name = string("op_1718_cast_fp16")]; bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1718_cast_fp16, y = obj_149_cast_fp16)[name = string("attn_31_cast_fp16")]; tensor var_1721 = const()[name = string("op_1721"), val = tensor([1, 768, 1, -1])]; tensor input_73_cast_fp16 = reshape(shape = var_1721, x = attn_31_cast_fp16)[name = string("input_73_cast_fp16")]; string obj_147_pad_type_0 = const()[name = string("obj_147_pad_type_0"), val = string("valid")]; tensor obj_147_strides_0 = const()[name = string("obj_147_strides_0"), val = tensor([1, 1])]; tensor obj_147_pad_0 = const()[name = string("obj_147_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_147_dilations_0 = const()[name = string("obj_147_dilations_0"), val = tensor([1, 1])]; int32 obj_147_groups_0 = const()[name = string("obj_147_groups_0"), val = int32(1)]; tensor layers_7_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202053184)))]; tensor layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203232896)))]; tensor obj_147_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_147_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_147_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_1742_to_fp16 = const()[name = string("op_1742_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1742_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_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(203234496)))]; 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(203236096)))]; 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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_47_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_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(203237696)))]; 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(207956352)))]; tensor input_77_cast_fp16 = conv(bias = layers_7_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_7_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_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_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(207962560)))]; 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(212681216)))]; tensor hidden_states_17_cast_fp16 = conv(bias = layers_7_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_7_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_49_cast_fp16")]; tensor obj_161_begin_0 = const()[name = string("obj_161_begin_0"), val = tensor([8, 0, 0, 0])]; tensor obj_161_end_0 = const()[name = string("obj_161_end_0"), val = tensor([9, 768, 1, 1536])]; tensor obj_161_end_mask_0 = const()[name = string("obj_161_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_161_cast_fp16 = slice_by_index(begin = obj_161_begin_0, end = obj_161_end_0, end_mask = obj_161_end_mask_0, x = read_state_2)[name = string("obj_161_cast_fp16")]; tensor obj_163_begin_0 = const()[name = string("obj_163_begin_0"), val = tensor([8, 0, 0, 0])]; tensor obj_163_end_0 = const()[name = string("obj_163_end_0"), val = tensor([9, 768, 1, 1536])]; tensor obj_163_end_mask_0 = const()[name = string("obj_163_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_163_cast_fp16 = slice_by_index(begin = obj_163_begin_0, end = obj_163_end_0, end_mask = obj_163_end_mask_0, x = read_state_3)[name = string("obj_163_cast_fp16")]; int32 var_1788 = const()[name = string("op_1788"), val = int32(3)]; tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; fp16 var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1813_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; tensor obj_151_gamma_0_to_fp16 = const()[name = string("obj_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212682816)))]; tensor obj_151_beta_0_to_fp16 = const()[name = string("obj_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212684416)))]; fp16 obj_151_epsilon_0_to_fp16 = const()[name = string("obj_151_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_151_cast_fp16 = batch_norm(beta = obj_151_beta_0_to_fp16, epsilon = obj_151_epsilon_0_to_fp16, gamma = obj_151_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_151_cast_fp16")]; string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; int32 query_33_groups_0 = const()[name = string("query_33_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(212686016)))]; 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(213865728)))]; tensor query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("query_33_cast_fp16")]; string current_key_17_pad_type_0 = const()[name = string("current_key_17_pad_type_0"), val = string("valid")]; tensor current_key_17_strides_0 = const()[name = string("current_key_17_strides_0"), val = tensor([1, 1])]; tensor current_key_17_pad_0 = const()[name = string("current_key_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_17_dilations_0 = const()[name = string("current_key_17_dilations_0"), val = tensor([1, 1])]; int32 current_key_17_groups_0 = const()[name = string("current_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(213867328)))]; tensor current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_key_17_cast_fp16")]; string current_value_17_pad_type_0 = const()[name = string("current_value_17_pad_type_0"), val = string("valid")]; tensor current_value_17_strides_0 = const()[name = string("current_value_17_strides_0"), val = tensor([1, 1])]; tensor current_value_17_pad_0 = const()[name = string("current_value_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_17_dilations_0 = const()[name = string("current_value_17_dilations_0"), val = tensor([1, 1])]; int32 current_value_17_groups_0 = const()[name = string("current_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(215047040)))]; 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(216226752)))]; tensor current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_value_17_cast_fp16")]; tensor var_1851_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1851_cast_fp16")]; tensor key_17_cast_fp16 = add(x = var_65_cast_fp16_8, y = var_1851_cast_fp16)[name = string("key_17_cast_fp16")]; tensor var_1853_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1853_cast_fp16")]; tensor value_17_cast_fp16 = add(x = var_80_cast_fp16_8, y = var_1853_cast_fp16)[name = string("value_17_cast_fp16")]; tensor var_1856 = const()[name = string("op_1856"), val = tensor([1, 12, 64, -1])]; tensor mh_q_33_cast_fp16 = reshape(shape = var_1856, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")]; fp16 var_1858_to_fp16 = const()[name = string("op_1858_to_fp16"), val = fp16(0x1p-3)]; tensor var_1859_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1858_to_fp16)[name = string("op_1859_cast_fp16")]; tensor var_1860 = const()[name = string("op_1860"), val = tensor([1, 12, 64, -1])]; tensor var_1861_cast_fp16 = reshape(shape = var_1860, x = key_17_cast_fp16)[name = string("op_1861_cast_fp16")]; bool mh_w_65_transpose_x_0 = const()[name = string("mh_w_65_transpose_x_0"), val = bool(true)]; bool mh_w_65_transpose_y_0 = const()[name = string("mh_w_65_transpose_y_0"), val = bool(false)]; tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_1859_cast_fp16, y = var_1861_cast_fp16)[name = string("mh_w_65_cast_fp16")]; tensor mh_w_67_cast_fp16 = add(x = mh_w_65_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_67_cast_fp16")]; tensor var_1869_cast_fp16 = softmax(axis = var_1788, x = mh_w_67_cast_fp16)[name = string("op_1869_cast_fp16")]; tensor var_1870 = const()[name = string("op_1870"), val = tensor([1, 12, 64, -1])]; tensor var_1871_cast_fp16 = reshape(shape = var_1870, x = value_17_cast_fp16)[name = string("op_1871_cast_fp16")]; bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1871_cast_fp16, y = var_1869_cast_fp16)[name = string("attn_33_cast_fp16")]; tensor var_1874 = const()[name = string("op_1874"), val = tensor([1, 768, 1, -1])]; tensor input_81_cast_fp16 = reshape(shape = var_1874, x = attn_33_cast_fp16)[name = string("input_81_cast_fp16")]; string obj_157_pad_type_0 = const()[name = string("obj_157_pad_type_0"), val = string("valid")]; tensor obj_157_strides_0 = const()[name = string("obj_157_strides_0"), val = tensor([1, 1])]; tensor obj_157_pad_0 = const()[name = string("obj_157_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_157_dilations_0 = const()[name = string("obj_157_dilations_0"), val = tensor([1, 1])]; int32 obj_157_groups_0 = const()[name = string("obj_157_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(216228352)))]; 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(217408064)))]; tensor obj_157_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_157_dilations_0, groups = obj_157_groups_0, pad = obj_157_pad_0, pad_type = obj_157_pad_type_0, strides = obj_157_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_157_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_157_cast_fp16)[name = string("inputs_51_cast_fp16")]; tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; fp16 var_1896_to_fp16 = const()[name = string("op_1896_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1896_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; tensor obj_159_gamma_0_to_fp16 = const()[name = string("obj_159_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217409664)))]; tensor obj_159_beta_0_to_fp16 = const()[name = string("obj_159_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217411264)))]; fp16 obj_159_epsilon_0_to_fp16 = const()[name = string("obj_159_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_159_cast_fp16 = batch_norm(beta = obj_159_beta_0_to_fp16, epsilon = obj_159_epsilon_0_to_fp16, gamma = obj_159_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("obj_159_cast_fp16")]; string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; tensor layers_8_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217412864)))]; tensor layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218592576)))]; tensor query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16, x = obj_159_cast_fp16)[name = string("query_35_cast_fp16")]; tensor var_1916 = const()[name = string("op_1916"), val = tensor([1, 12, 64, -1])]; tensor mh_q_35_cast_fp16 = reshape(shape = var_1916, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")]; fp16 var_1918_to_fp16 = const()[name = string("op_1918_to_fp16"), val = fp16(0x1p-3)]; tensor var_1919_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_1918_to_fp16)[name = string("op_1919_cast_fp16")]; tensor var_1920 = const()[name = string("op_1920"), val = tensor([1, 12, 64, -1])]; tensor var_1921_cast_fp16 = reshape(shape = var_1920, x = obj_161_cast_fp16)[name = string("op_1921_cast_fp16")]; bool mh_w_69_transpose_x_0 = const()[name = string("mh_w_69_transpose_x_0"), val = bool(true)]; bool mh_w_69_transpose_y_0 = const()[name = string("mh_w_69_transpose_y_0"), val = bool(false)]; tensor mh_w_69_cast_fp16 = matmul(transpose_x = mh_w_69_transpose_x_0, transpose_y = mh_w_69_transpose_y_0, x = var_1919_cast_fp16, y = var_1921_cast_fp16)[name = string("mh_w_69_cast_fp16")]; tensor mh_w_71_cast_fp16 = add(x = mh_w_69_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_71_cast_fp16")]; tensor obj_167_cast_fp16 = softmax(axis = var_1788, x = mh_w_71_cast_fp16)[name = string("obj_167_cast_fp16")]; tensor var_1930 = const()[name = string("op_1930"), val = tensor([1, 12, 64, -1])]; tensor var_1931_cast_fp16 = reshape(shape = var_1930, x = obj_163_cast_fp16)[name = string("op_1931_cast_fp16")]; bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_1931_cast_fp16, y = obj_167_cast_fp16)[name = string("attn_35_cast_fp16")]; tensor var_1934 = const()[name = string("op_1934"), val = tensor([1, 768, 1, -1])]; tensor input_83_cast_fp16 = reshape(shape = var_1934, x = attn_35_cast_fp16)[name = string("input_83_cast_fp16")]; string obj_165_pad_type_0 = const()[name = string("obj_165_pad_type_0"), val = string("valid")]; tensor obj_165_strides_0 = const()[name = string("obj_165_strides_0"), val = tensor([1, 1])]; tensor obj_165_pad_0 = const()[name = string("obj_165_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_165_dilations_0 = const()[name = string("obj_165_dilations_0"), val = tensor([1, 1])]; int32 obj_165_groups_0 = const()[name = string("obj_165_groups_0"), val = int32(1)]; tensor layers_8_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218594176)))]; tensor layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219773888)))]; tensor obj_165_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("obj_165_cast_fp16")]; tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_165_cast_fp16)[name = string("inputs_53_cast_fp16")]; tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; fp16 var_1955_to_fp16 = const()[name = string("op_1955_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1955_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; tensor input_85_gamma_0_to_fp16 = const()[name = string("input_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219775488)))]; tensor input_85_beta_0_to_fp16 = const()[name = string("input_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219777088)))]; fp16 input_85_epsilon_0_to_fp16 = const()[name = string("input_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("input_85_cast_fp16")]; string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("valid")]; tensor input_87_strides_0 = const()[name = string("input_87_strides_0"), val = tensor([1, 1])]; tensor input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_87_dilations_0 = const()[name = string("input_87_dilations_0"), val = tensor([1, 1])]; int32 input_87_groups_0 = const()[name = string("input_87_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(219778688)))]; 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(224497344)))]; tensor input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; string input_89_mode_0 = const()[name = string("input_89_mode_0"), val = string("EXACT")]; tensor input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = string("input_89_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_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(224503552)))]; 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(229222208)))]; tensor hidden_states_19_cast_fp16 = conv(bias = layers_8_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_8_fc2_weight_to_fp16, x = input_89_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_55_cast_fp16")]; tensor obj_179_begin_0 = const()[name = string("obj_179_begin_0"), val = tensor([9, 0, 0, 0])]; tensor obj_179_end_0 = const()[name = string("obj_179_end_0"), val = tensor([10, 768, 1, 1536])]; tensor obj_179_end_mask_0 = const()[name = string("obj_179_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_179_cast_fp16 = slice_by_index(begin = obj_179_begin_0, end = obj_179_end_0, end_mask = obj_179_end_mask_0, x = read_state_2)[name = string("obj_179_cast_fp16")]; tensor obj_181_begin_0 = const()[name = string("obj_181_begin_0"), val = tensor([9, 0, 0, 0])]; tensor obj_181_end_0 = const()[name = string("obj_181_end_0"), val = tensor([10, 768, 1, 1536])]; tensor obj_181_end_mask_0 = const()[name = string("obj_181_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_181_cast_fp16 = slice_by_index(begin = obj_181_begin_0, end = obj_181_end_0, end_mask = obj_181_end_mask_0, x = read_state_3)[name = string("obj_181_cast_fp16")]; int32 var_2001 = const()[name = string("op_2001"), val = int32(3)]; tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; fp16 var_2026_to_fp16 = const()[name = string("op_2026_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2026_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; tensor obj_169_gamma_0_to_fp16 = const()[name = string("obj_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229223808)))]; tensor obj_169_beta_0_to_fp16 = const()[name = string("obj_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229225408)))]; fp16 obj_169_epsilon_0_to_fp16 = const()[name = string("obj_169_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("obj_169_cast_fp16")]; string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; int32 query_37_groups_0 = const()[name = string("query_37_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(229227008)))]; 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(230406720)))]; tensor query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("query_37_cast_fp16")]; string current_key_19_pad_type_0 = const()[name = string("current_key_19_pad_type_0"), val = string("valid")]; tensor current_key_19_strides_0 = const()[name = string("current_key_19_strides_0"), val = tensor([1, 1])]; tensor current_key_19_pad_0 = const()[name = string("current_key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_19_dilations_0 = const()[name = string("current_key_19_dilations_0"), val = tensor([1, 1])]; int32 current_key_19_groups_0 = const()[name = string("current_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(230408320)))]; tensor current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_key_19_cast_fp16")]; string current_value_19_pad_type_0 = const()[name = string("current_value_19_pad_type_0"), val = string("valid")]; tensor current_value_19_strides_0 = const()[name = string("current_value_19_strides_0"), val = tensor([1, 1])]; tensor current_value_19_pad_0 = const()[name = string("current_value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_19_dilations_0 = const()[name = string("current_value_19_dilations_0"), val = tensor([1, 1])]; int32 current_value_19_groups_0 = const()[name = string("current_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(231588032)))]; 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(232767744)))]; tensor current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_value_19_cast_fp16")]; tensor var_2064_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2064_cast_fp16")]; tensor key_19_cast_fp16 = add(x = var_65_cast_fp16_9, y = var_2064_cast_fp16)[name = string("key_19_cast_fp16")]; tensor var_2066_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2066_cast_fp16")]; tensor value_19_cast_fp16 = add(x = var_80_cast_fp16_9, y = var_2066_cast_fp16)[name = string("value_19_cast_fp16")]; tensor var_2069 = const()[name = string("op_2069"), val = tensor([1, 12, 64, -1])]; tensor mh_q_37_cast_fp16 = reshape(shape = var_2069, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")]; fp16 var_2071_to_fp16 = const()[name = string("op_2071_to_fp16"), val = fp16(0x1p-3)]; tensor var_2072_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2071_to_fp16)[name = string("op_2072_cast_fp16")]; tensor var_2073 = const()[name = string("op_2073"), val = tensor([1, 12, 64, -1])]; tensor var_2074_cast_fp16 = reshape(shape = var_2073, x = key_19_cast_fp16)[name = string("op_2074_cast_fp16")]; bool mh_w_73_transpose_x_0 = const()[name = string("mh_w_73_transpose_x_0"), val = bool(true)]; bool mh_w_73_transpose_y_0 = const()[name = string("mh_w_73_transpose_y_0"), val = bool(false)]; tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2072_cast_fp16, y = var_2074_cast_fp16)[name = string("mh_w_73_cast_fp16")]; tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_75_cast_fp16")]; tensor var_2082_cast_fp16 = softmax(axis = var_2001, x = mh_w_75_cast_fp16)[name = string("op_2082_cast_fp16")]; tensor var_2083 = const()[name = string("op_2083"), val = tensor([1, 12, 64, -1])]; tensor var_2084_cast_fp16 = reshape(shape = var_2083, x = value_19_cast_fp16)[name = string("op_2084_cast_fp16")]; bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2084_cast_fp16, y = var_2082_cast_fp16)[name = string("attn_37_cast_fp16")]; tensor var_2087 = const()[name = string("op_2087"), val = tensor([1, 768, 1, -1])]; tensor input_91_cast_fp16 = reshape(shape = var_2087, x = attn_37_cast_fp16)[name = string("input_91_cast_fp16")]; string obj_175_pad_type_0 = const()[name = string("obj_175_pad_type_0"), val = string("valid")]; tensor obj_175_strides_0 = const()[name = string("obj_175_strides_0"), val = tensor([1, 1])]; tensor obj_175_pad_0 = const()[name = string("obj_175_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_175_dilations_0 = const()[name = string("obj_175_dilations_0"), val = tensor([1, 1])]; int32 obj_175_groups_0 = const()[name = string("obj_175_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(232769344)))]; 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(233949056)))]; tensor obj_175_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("obj_175_cast_fp16")]; tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_175_cast_fp16)[name = string("inputs_57_cast_fp16")]; tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; fp16 var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2109_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; tensor obj_177_gamma_0_to_fp16 = const()[name = string("obj_177_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233950656)))]; tensor obj_177_beta_0_to_fp16 = const()[name = string("obj_177_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233952256)))]; fp16 obj_177_epsilon_0_to_fp16 = const()[name = string("obj_177_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_177_cast_fp16")]; string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; tensor layers_9_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233953856)))]; tensor layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235133568)))]; tensor query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16, x = obj_177_cast_fp16)[name = string("query_39_cast_fp16")]; tensor var_2129 = const()[name = string("op_2129"), val = tensor([1, 12, 64, -1])]; tensor mh_q_39_cast_fp16 = reshape(shape = var_2129, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")]; fp16 var_2131_to_fp16 = const()[name = string("op_2131_to_fp16"), val = fp16(0x1p-3)]; tensor var_2132_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2131_to_fp16)[name = string("op_2132_cast_fp16")]; tensor var_2133 = const()[name = string("op_2133"), val = tensor([1, 12, 64, -1])]; tensor var_2134_cast_fp16 = reshape(shape = var_2133, x = obj_179_cast_fp16)[name = string("op_2134_cast_fp16")]; bool mh_w_77_transpose_x_0 = const()[name = string("mh_w_77_transpose_x_0"), val = bool(true)]; bool mh_w_77_transpose_y_0 = const()[name = string("mh_w_77_transpose_y_0"), val = bool(false)]; tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2132_cast_fp16, y = var_2134_cast_fp16)[name = string("mh_w_77_cast_fp16")]; tensor mh_w_79_cast_fp16 = add(x = mh_w_77_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_79_cast_fp16")]; tensor obj_185_cast_fp16 = softmax(axis = var_2001, x = mh_w_79_cast_fp16)[name = string("obj_185_cast_fp16")]; tensor var_2143 = const()[name = string("op_2143"), val = tensor([1, 12, 64, -1])]; tensor var_2144_cast_fp16 = reshape(shape = var_2143, x = obj_181_cast_fp16)[name = string("op_2144_cast_fp16")]; bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2144_cast_fp16, y = obj_185_cast_fp16)[name = string("attn_39_cast_fp16")]; tensor var_2147 = const()[name = string("op_2147"), val = tensor([1, 768, 1, -1])]; tensor input_93_cast_fp16 = reshape(shape = var_2147, x = attn_39_cast_fp16)[name = string("input_93_cast_fp16")]; string obj_183_pad_type_0 = const()[name = string("obj_183_pad_type_0"), val = string("valid")]; tensor obj_183_strides_0 = const()[name = string("obj_183_strides_0"), val = tensor([1, 1])]; tensor obj_183_pad_0 = const()[name = string("obj_183_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_183_dilations_0 = const()[name = string("obj_183_dilations_0"), val = tensor([1, 1])]; int32 obj_183_groups_0 = const()[name = string("obj_183_groups_0"), val = int32(1)]; tensor layers_9_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235135168)))]; tensor layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236314880)))]; tensor obj_183_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_183_dilations_0, groups = obj_183_groups_0, pad = obj_183_pad_0, pad_type = obj_183_pad_type_0, strides = obj_183_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("obj_183_cast_fp16")]; tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_183_cast_fp16)[name = string("inputs_59_cast_fp16")]; tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; fp16 var_2168_to_fp16 = const()[name = string("op_2168_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2168_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; tensor input_95_gamma_0_to_fp16 = const()[name = string("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236316480)))]; tensor input_95_beta_0_to_fp16 = const()[name = string("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236318080)))]; fp16 input_95_epsilon_0_to_fp16 = const()[name = string("input_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_95_cast_fp16")]; string input_97_pad_type_0 = const()[name = string("input_97_pad_type_0"), val = string("valid")]; tensor input_97_strides_0 = const()[name = string("input_97_strides_0"), val = tensor([1, 1])]; tensor input_97_pad_0 = const()[name = string("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_97_dilations_0 = const()[name = string("input_97_dilations_0"), val = tensor([1, 1])]; int32 input_97_groups_0 = const()[name = string("input_97_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(236319680)))]; 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(241038336)))]; tensor input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_95_cast_fp16)[name = string("input_97_cast_fp16")]; string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")]; tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = string("input_99_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_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(241044544)))]; 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(245763200)))]; tensor hidden_states_21_cast_fp16 = conv(bias = layers_9_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_9_fc2_weight_to_fp16, x = input_99_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_61_cast_fp16")]; tensor obj_197_begin_0 = const()[name = string("obj_197_begin_0"), val = tensor([10, 0, 0, 0])]; tensor obj_197_end_0 = const()[name = string("obj_197_end_0"), val = tensor([11, 768, 1, 1536])]; tensor obj_197_end_mask_0 = const()[name = string("obj_197_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_197_cast_fp16 = slice_by_index(begin = obj_197_begin_0, end = obj_197_end_0, end_mask = obj_197_end_mask_0, x = read_state_2)[name = string("obj_197_cast_fp16")]; tensor obj_199_begin_0 = const()[name = string("obj_199_begin_0"), val = tensor([10, 0, 0, 0])]; tensor obj_199_end_0 = const()[name = string("obj_199_end_0"), val = tensor([11, 768, 1, 1536])]; tensor obj_199_end_mask_0 = const()[name = string("obj_199_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_199_cast_fp16 = slice_by_index(begin = obj_199_begin_0, end = obj_199_end_0, end_mask = obj_199_end_mask_0, x = read_state_3)[name = string("obj_199_cast_fp16")]; int32 var_2214 = const()[name = string("op_2214"), val = int32(3)]; tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; fp16 var_2239_to_fp16 = const()[name = string("op_2239_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2239_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; tensor obj_187_gamma_0_to_fp16 = const()[name = string("obj_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245764800)))]; tensor obj_187_beta_0_to_fp16 = const()[name = string("obj_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245766400)))]; fp16 obj_187_epsilon_0_to_fp16 = const()[name = string("obj_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_187_cast_fp16 = batch_norm(beta = obj_187_beta_0_to_fp16, epsilon = obj_187_epsilon_0_to_fp16, gamma = obj_187_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_187_cast_fp16")]; string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; int32 query_41_groups_0 = const()[name = string("query_41_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(245768000)))]; 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(246947712)))]; tensor query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("query_41_cast_fp16")]; string current_key_21_pad_type_0 = const()[name = string("current_key_21_pad_type_0"), val = string("valid")]; tensor current_key_21_strides_0 = const()[name = string("current_key_21_strides_0"), val = tensor([1, 1])]; tensor current_key_21_pad_0 = const()[name = string("current_key_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_21_dilations_0 = const()[name = string("current_key_21_dilations_0"), val = tensor([1, 1])]; int32 current_key_21_groups_0 = const()[name = string("current_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(246949312)))]; tensor current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_key_21_cast_fp16")]; string current_value_21_pad_type_0 = const()[name = string("current_value_21_pad_type_0"), val = string("valid")]; tensor current_value_21_strides_0 = const()[name = string("current_value_21_strides_0"), val = tensor([1, 1])]; tensor current_value_21_pad_0 = const()[name = string("current_value_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_21_dilations_0 = const()[name = string("current_value_21_dilations_0"), val = tensor([1, 1])]; int32 current_value_21_groups_0 = const()[name = string("current_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(248129024)))]; 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(249308736)))]; tensor current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_value_21_cast_fp16")]; tensor var_2277_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2277_cast_fp16")]; tensor key_21_cast_fp16 = add(x = var_65_cast_fp16_10, y = var_2277_cast_fp16)[name = string("key_21_cast_fp16")]; tensor var_2279_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2279_cast_fp16")]; tensor value_21_cast_fp16 = add(x = var_80_cast_fp16_10, y = var_2279_cast_fp16)[name = string("value_21_cast_fp16")]; tensor var_2282 = const()[name = string("op_2282"), val = tensor([1, 12, 64, -1])]; tensor mh_q_41_cast_fp16 = reshape(shape = var_2282, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")]; fp16 var_2284_to_fp16 = const()[name = string("op_2284_to_fp16"), val = fp16(0x1p-3)]; tensor var_2285_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2284_to_fp16)[name = string("op_2285_cast_fp16")]; tensor var_2286 = const()[name = string("op_2286"), val = tensor([1, 12, 64, -1])]; tensor var_2287_cast_fp16 = reshape(shape = var_2286, x = key_21_cast_fp16)[name = string("op_2287_cast_fp16")]; bool mh_w_81_transpose_x_0 = const()[name = string("mh_w_81_transpose_x_0"), val = bool(true)]; bool mh_w_81_transpose_y_0 = const()[name = string("mh_w_81_transpose_y_0"), val = bool(false)]; tensor mh_w_81_cast_fp16 = matmul(transpose_x = mh_w_81_transpose_x_0, transpose_y = mh_w_81_transpose_y_0, x = var_2285_cast_fp16, y = var_2287_cast_fp16)[name = string("mh_w_81_cast_fp16")]; tensor mh_w_83_cast_fp16 = add(x = mh_w_81_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_83_cast_fp16")]; tensor var_2295_cast_fp16 = softmax(axis = var_2214, x = mh_w_83_cast_fp16)[name = string("op_2295_cast_fp16")]; tensor var_2296 = const()[name = string("op_2296"), val = tensor([1, 12, 64, -1])]; tensor var_2297_cast_fp16 = reshape(shape = var_2296, x = value_21_cast_fp16)[name = string("op_2297_cast_fp16")]; bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2297_cast_fp16, y = var_2295_cast_fp16)[name = string("attn_41_cast_fp16")]; tensor var_2300 = const()[name = string("op_2300"), val = tensor([1, 768, 1, -1])]; tensor input_101_cast_fp16 = reshape(shape = var_2300, x = attn_41_cast_fp16)[name = string("input_101_cast_fp16")]; string obj_193_pad_type_0 = const()[name = string("obj_193_pad_type_0"), val = string("valid")]; tensor obj_193_strides_0 = const()[name = string("obj_193_strides_0"), val = tensor([1, 1])]; tensor obj_193_pad_0 = const()[name = string("obj_193_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_193_dilations_0 = const()[name = string("obj_193_dilations_0"), val = tensor([1, 1])]; int32 obj_193_groups_0 = const()[name = string("obj_193_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(249310336)))]; 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(250490048)))]; tensor obj_193_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_101_cast_fp16)[name = string("obj_193_cast_fp16")]; tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_193_cast_fp16)[name = string("inputs_63_cast_fp16")]; tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2322_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; tensor obj_195_gamma_0_to_fp16 = const()[name = string("obj_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250491648)))]; tensor obj_195_beta_0_to_fp16 = const()[name = string("obj_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250493248)))]; fp16 obj_195_epsilon_0_to_fp16 = const()[name = string("obj_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_195_cast_fp16 = batch_norm(beta = obj_195_beta_0_to_fp16, epsilon = obj_195_epsilon_0_to_fp16, gamma = obj_195_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("obj_195_cast_fp16")]; string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; tensor layers_10_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250494848)))]; tensor layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251674560)))]; tensor query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16, x = obj_195_cast_fp16)[name = string("query_43_cast_fp16")]; tensor var_2342 = const()[name = string("op_2342"), val = tensor([1, 12, 64, -1])]; tensor mh_q_43_cast_fp16 = reshape(shape = var_2342, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")]; fp16 var_2344_to_fp16 = const()[name = string("op_2344_to_fp16"), val = fp16(0x1p-3)]; tensor var_2345_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2344_to_fp16)[name = string("op_2345_cast_fp16")]; tensor var_2346 = const()[name = string("op_2346"), val = tensor([1, 12, 64, -1])]; tensor var_2347_cast_fp16 = reshape(shape = var_2346, x = obj_197_cast_fp16)[name = string("op_2347_cast_fp16")]; bool mh_w_85_transpose_x_0 = const()[name = string("mh_w_85_transpose_x_0"), val = bool(true)]; bool mh_w_85_transpose_y_0 = const()[name = string("mh_w_85_transpose_y_0"), val = bool(false)]; tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_2345_cast_fp16, y = var_2347_cast_fp16)[name = string("mh_w_85_cast_fp16")]; tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_87_cast_fp16")]; tensor obj_203_cast_fp16 = softmax(axis = var_2214, x = mh_w_87_cast_fp16)[name = string("obj_203_cast_fp16")]; tensor var_2356 = const()[name = string("op_2356"), val = tensor([1, 12, 64, -1])]; tensor var_2357_cast_fp16 = reshape(shape = var_2356, x = obj_199_cast_fp16)[name = string("op_2357_cast_fp16")]; bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2357_cast_fp16, y = obj_203_cast_fp16)[name = string("attn_43_cast_fp16")]; tensor var_2360 = const()[name = string("op_2360"), val = tensor([1, 768, 1, -1])]; tensor input_103_cast_fp16 = reshape(shape = var_2360, x = attn_43_cast_fp16)[name = string("input_103_cast_fp16")]; string obj_201_pad_type_0 = const()[name = string("obj_201_pad_type_0"), val = string("valid")]; tensor obj_201_strides_0 = const()[name = string("obj_201_strides_0"), val = tensor([1, 1])]; tensor obj_201_pad_0 = const()[name = string("obj_201_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_201_dilations_0 = const()[name = string("obj_201_dilations_0"), val = tensor([1, 1])]; int32 obj_201_groups_0 = const()[name = string("obj_201_groups_0"), val = int32(1)]; tensor layers_10_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251676160)))]; tensor layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252855872)))]; tensor obj_201_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_201_dilations_0, groups = obj_201_groups_0, pad = obj_201_pad_0, pad_type = obj_201_pad_type_0, strides = obj_201_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("obj_201_cast_fp16")]; tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_201_cast_fp16)[name = string("inputs_65_cast_fp16")]; tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; fp16 var_2381_to_fp16 = const()[name = string("op_2381_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2381_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; tensor input_105_gamma_0_to_fp16 = const()[name = string("input_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252857472)))]; tensor input_105_beta_0_to_fp16 = const()[name = string("input_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252859072)))]; fp16 input_105_epsilon_0_to_fp16 = const()[name = string("input_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("input_105_cast_fp16")]; string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")]; tensor input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor([1, 1])]; tensor input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor([1, 1])]; int32 input_107_groups_0 = const()[name = string("input_107_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(252860672)))]; 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(257579328)))]; tensor input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")]; string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")]; tensor input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = string("input_109_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_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(257585536)))]; 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(262304192)))]; tensor hidden_states_23_cast_fp16 = conv(bias = layers_10_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_10_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_67_cast_fp16")]; tensor obj_215_begin_0 = const()[name = string("obj_215_begin_0"), val = tensor([11, 0, 0, 0])]; tensor obj_215_end_0 = const()[name = string("obj_215_end_0"), val = tensor([12, 768, 1, 1536])]; tensor obj_215_end_mask_0 = const()[name = string("obj_215_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_215_cast_fp16 = slice_by_index(begin = obj_215_begin_0, end = obj_215_end_0, end_mask = obj_215_end_mask_0, x = read_state_2)[name = string("obj_215_cast_fp16")]; tensor obj_217_begin_0 = const()[name = string("obj_217_begin_0"), val = tensor([11, 0, 0, 0])]; tensor obj_217_end_0 = const()[name = string("obj_217_end_0"), val = tensor([12, 768, 1, 1536])]; tensor obj_217_end_mask_0 = const()[name = string("obj_217_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_217_cast_fp16 = slice_by_index(begin = obj_217_begin_0, end = obj_217_end_0, end_mask = obj_217_end_mask_0, x = read_state_3)[name = string("obj_217_cast_fp16")]; int32 var_2427 = const()[name = string("op_2427"), val = int32(3)]; tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; fp16 var_2452_to_fp16 = const()[name = string("op_2452_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2452_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; tensor obj_205_gamma_0_to_fp16 = const()[name = string("obj_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262305792)))]; tensor obj_205_beta_0_to_fp16 = const()[name = string("obj_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262307392)))]; fp16 obj_205_epsilon_0_to_fp16 = const()[name = string("obj_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("obj_205_cast_fp16")]; string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; int32 query_45_groups_0 = const()[name = string("query_45_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(262308992)))]; 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(263488704)))]; tensor query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("query_45_cast_fp16")]; string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; int32 current_key_groups_0 = const()[name = string("current_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(263490304)))]; tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_key_cast_fp16")]; string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; int32 current_value_groups_0 = const()[name = string("current_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(264670016)))]; 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(265849728)))]; tensor current_value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_value_cast_fp16")]; tensor var_2490_cast_fp16 = mul(x = current_key_cast_fp16, y = var_169_cast_fp16)[name = string("op_2490_cast_fp16")]; tensor key_cast_fp16 = add(x = var_65_cast_fp16_11, y = var_2490_cast_fp16)[name = string("key_cast_fp16")]; tensor var_2492_cast_fp16 = mul(x = current_value_cast_fp16, y = var_169_cast_fp16)[name = string("op_2492_cast_fp16")]; tensor value_cast_fp16 = add(x = var_80_cast_fp16_11, y = var_2492_cast_fp16)[name = string("value_cast_fp16")]; tensor var_2495 = const()[name = string("op_2495"), val = tensor([1, 12, 64, -1])]; tensor mh_q_45_cast_fp16 = reshape(shape = var_2495, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")]; fp16 var_2497_to_fp16 = const()[name = string("op_2497_to_fp16"), val = fp16(0x1p-3)]; tensor var_2498_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2497_to_fp16)[name = string("op_2498_cast_fp16")]; tensor var_2499 = const()[name = string("op_2499"), val = tensor([1, 12, 64, -1])]; tensor var_2500_cast_fp16 = reshape(shape = var_2499, x = key_cast_fp16)[name = string("op_2500_cast_fp16")]; bool mh_w_89_transpose_x_0 = const()[name = string("mh_w_89_transpose_x_0"), val = bool(true)]; bool mh_w_89_transpose_y_0 = const()[name = string("mh_w_89_transpose_y_0"), val = bool(false)]; tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_2498_cast_fp16, y = var_2500_cast_fp16)[name = string("mh_w_89_cast_fp16")]; tensor mh_w_91_cast_fp16 = add(x = mh_w_89_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_91_cast_fp16")]; tensor var_2508_cast_fp16 = softmax(axis = var_2427, x = mh_w_91_cast_fp16)[name = string("op_2508_cast_fp16")]; tensor var_2509 = const()[name = string("op_2509"), val = tensor([1, 12, 64, -1])]; tensor var_2510_cast_fp16 = reshape(shape = var_2509, x = value_cast_fp16)[name = string("op_2510_cast_fp16")]; bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2510_cast_fp16, y = var_2508_cast_fp16)[name = string("attn_45_cast_fp16")]; tensor var_2513 = const()[name = string("op_2513"), val = tensor([1, 768, 1, -1])]; tensor input_111_cast_fp16 = reshape(shape = var_2513, x = attn_45_cast_fp16)[name = string("input_111_cast_fp16")]; string obj_211_pad_type_0 = const()[name = string("obj_211_pad_type_0"), val = string("valid")]; tensor obj_211_strides_0 = const()[name = string("obj_211_strides_0"), val = tensor([1, 1])]; tensor obj_211_pad_0 = const()[name = string("obj_211_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_211_dilations_0 = const()[name = string("obj_211_dilations_0"), val = tensor([1, 1])]; int32 obj_211_groups_0 = const()[name = string("obj_211_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(265851328)))]; 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(267031040)))]; tensor obj_211_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_211_dilations_0, groups = obj_211_groups_0, pad = obj_211_pad_0, pad_type = obj_211_pad_type_0, strides = obj_211_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("obj_211_cast_fp16")]; tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_211_cast_fp16)[name = string("inputs_69_cast_fp16")]; tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; fp16 var_2535_to_fp16 = const()[name = string("op_2535_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2535_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; tensor obj_213_gamma_0_to_fp16 = const()[name = string("obj_213_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267032640)))]; tensor obj_213_beta_0_to_fp16 = const()[name = string("obj_213_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267034240)))]; fp16 obj_213_epsilon_0_to_fp16 = const()[name = string("obj_213_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_213_cast_fp16 = batch_norm(beta = obj_213_beta_0_to_fp16, epsilon = obj_213_epsilon_0_to_fp16, gamma = obj_213_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_213_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_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267035840)))]; tensor layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268215552)))]; tensor query_cast_fp16 = conv(bias = layers_11_encoder_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_encoder_attn_q_proj_weight_to_fp16, x = obj_213_cast_fp16)[name = string("query_cast_fp16")]; tensor var_2555 = const()[name = string("op_2555"), val = tensor([1, 12, 64, -1])]; tensor mh_q_cast_fp16 = reshape(shape = var_2555, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; fp16 var_2557_to_fp16 = const()[name = string("op_2557_to_fp16"), val = fp16(0x1p-3)]; tensor var_2558_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2557_to_fp16)[name = string("op_2558_cast_fp16")]; tensor var_2559 = const()[name = string("op_2559"), val = tensor([1, 12, 64, -1])]; tensor var_2560_cast_fp16 = reshape(shape = var_2559, x = obj_215_cast_fp16)[name = string("op_2560_cast_fp16")]; bool mh_w_93_transpose_x_0 = const()[name = string("mh_w_93_transpose_x_0"), val = bool(true)]; bool mh_w_93_transpose_y_0 = const()[name = string("mh_w_93_transpose_y_0"), val = bool(false)]; tensor mh_w_93_cast_fp16 = matmul(transpose_x = mh_w_93_transpose_x_0, transpose_y = mh_w_93_transpose_y_0, x = var_2558_cast_fp16, y = var_2560_cast_fp16)[name = string("mh_w_93_cast_fp16")]; tensor mh_w_cast_fp16 = add(x = mh_w_93_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_cast_fp16")]; tensor obj_221_cast_fp16 = softmax(axis = var_2427, x = mh_w_cast_fp16)[name = string("obj_221_cast_fp16")]; tensor var_2569 = const()[name = string("op_2569"), val = tensor([1, 12, 64, -1])]; tensor var_2570_cast_fp16 = reshape(shape = var_2569, x = obj_217_cast_fp16)[name = string("op_2570_cast_fp16")]; bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2570_cast_fp16, y = obj_221_cast_fp16)[name = string("attn_cast_fp16")]; tensor var_2573 = const()[name = string("op_2573"), val = tensor([1, 768, 1, -1])]; tensor input_113_cast_fp16 = reshape(shape = var_2573, x = attn_cast_fp16)[name = string("input_113_cast_fp16")]; string obj_219_pad_type_0 = const()[name = string("obj_219_pad_type_0"), val = string("valid")]; tensor obj_219_strides_0 = const()[name = string("obj_219_strides_0"), val = tensor([1, 1])]; tensor obj_219_pad_0 = const()[name = string("obj_219_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_219_dilations_0 = const()[name = string("obj_219_dilations_0"), val = tensor([1, 1])]; int32 obj_219_groups_0 = const()[name = string("obj_219_groups_0"), val = int32(1)]; tensor layers_11_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268217152)))]; tensor layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269396864)))]; tensor obj_219_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_219_dilations_0, groups = obj_219_groups_0, pad = obj_219_pad_0, pad_type = obj_219_pad_type_0, strides = obj_219_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_219_cast_fp16")]; tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_219_cast_fp16)[name = string("inputs_71_cast_fp16")]; tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; fp16 var_2594_to_fp16 = const()[name = string("op_2594_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2594_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269398464)))]; tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269400064)))]; fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_115_cast_fp16")]; string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; int32 input_117_groups_0 = const()[name = string("input_117_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(269401664)))]; 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(274120320)))]; tensor input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_117_cast_fp16)[name = string("input_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_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(274126528)))]; 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(278845184)))]; tensor hidden_states_25_cast_fp16 = conv(bias = layers_11_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_11_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; tensor inputs_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_cast_fp16")]; tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; fp16 var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2637_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278846784)))]; tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278848384)))]; fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; tensor var_2648_axes_0 = const()[name = string("op_2648_axes_0"), val = tensor([2])]; tensor var_2648_cast_fp16 = squeeze(axes = var_2648_axes_0, x = hidden_states_cast_fp16)[name = string("op_2648_cast_fp16")]; tensor var_2651_perm_0 = const()[name = string("op_2651_perm_0"), val = tensor([0, 2, 1])]; tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278849984)))]; tensor var_2651_cast_fp16 = transpose(perm = var_2651_perm_0, x = var_2648_cast_fp16)[name = string("transpose_0")]; tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_2651_cast_fp16)[name = string("linear_0_cast_fp16")]; int32 var_2655 = const()[name = string("op_2655"), val = int32(1)]; bool obj_225_interleave_0 = const()[name = string("obj_225_interleave_0"), val = bool(false)]; tensor key_cache_updates = concat(axis = var_2655, interleave = obj_225_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_cast_fp16))[name = string("obj_225_cast_fp16")]; int32 var_2658 = const()[name = string("op_2658"), val = int32(1)]; bool obj_227_interleave_0 = const()[name = string("obj_227_interleave_0"), val = bool(false)]; tensor value_cache_updates = concat(axis = var_2658, interleave = obj_227_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_cast_fp16))[name = string("obj_227_cast_fp16")]; tensor var_2669_begin_0 = const()[name = string("op_2669_begin_0"), val = tensor([0, 6, 0, 0])]; tensor var_2669_end_0 = const()[name = string("op_2669_end_0"), val = tensor([1, 7, 1, 1536])]; tensor var_2669_end_mask_0 = const()[name = string("op_2669_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2669_cast_fp16 = slice_by_index(begin = var_2669_begin_0, end = var_2669_end_0, end_mask = var_2669_end_mask_0, x = obj_131_cast_fp16)[name = string("op_2669_cast_fp16")]; tensor var_2672_begin_0 = const()[name = string("op_2672_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2672_end_0 = const()[name = string("op_2672_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2672_end_mask_0 = const()[name = string("op_2672_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2672_squeeze_mask_0 = const()[name = string("op_2672_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2672_cast_fp16 = slice_by_index(begin = var_2672_begin_0, end = var_2672_end_0, end_mask = var_2672_end_mask_0, squeeze_mask = var_2672_squeeze_mask_0, x = var_2669_cast_fp16)[name = string("op_2672_cast_fp16")]; tensor var_2687_begin_0 = const()[name = string("op_2687_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2687_end_0 = const()[name = string("op_2687_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2687_end_mask_0 = const()[name = string("op_2687_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2687_cast_fp16 = slice_by_index(begin = var_2687_begin_0, end = var_2687_end_0, end_mask = var_2687_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2687_cast_fp16")]; tensor var_2690_begin_0 = const()[name = string("op_2690_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2690_end_0 = const()[name = string("op_2690_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2690_end_mask_0 = const()[name = string("op_2690_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2690_squeeze_mask_0 = const()[name = string("op_2690_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2690_cast_fp16 = slice_by_index(begin = var_2690_begin_0, end = var_2690_end_0, end_mask = var_2690_end_mask_0, squeeze_mask = var_2690_squeeze_mask_0, x = var_2687_cast_fp16)[name = string("op_2690_cast_fp16")]; tensor var_2705_begin_0 = const()[name = string("op_2705_begin_0"), val = tensor([0, 3, 0, 0])]; tensor var_2705_end_0 = const()[name = string("op_2705_end_0"), val = tensor([1, 4, 1, 1536])]; tensor var_2705_end_mask_0 = const()[name = string("op_2705_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2705_cast_fp16 = slice_by_index(begin = var_2705_begin_0, end = var_2705_end_0, end_mask = var_2705_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2705_cast_fp16")]; tensor var_2708_begin_0 = const()[name = string("op_2708_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2708_end_0 = const()[name = string("op_2708_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2708_end_mask_0 = const()[name = string("op_2708_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2708_squeeze_mask_0 = const()[name = string("op_2708_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2708_cast_fp16 = slice_by_index(begin = var_2708_begin_0, end = var_2708_end_0, end_mask = var_2708_end_mask_0, squeeze_mask = var_2708_squeeze_mask_0, x = var_2705_cast_fp16)[name = string("op_2708_cast_fp16")]; tensor var_2723_begin_0 = const()[name = string("op_2723_begin_0"), val = tensor([0, 8, 0, 0])]; tensor var_2723_end_0 = const()[name = string("op_2723_end_0"), val = tensor([1, 9, 1, 1536])]; tensor var_2723_end_mask_0 = const()[name = string("op_2723_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2723_cast_fp16 = slice_by_index(begin = var_2723_begin_0, end = var_2723_end_0, end_mask = var_2723_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2723_cast_fp16")]; tensor var_2726_begin_0 = const()[name = string("op_2726_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2726_end_0 = const()[name = string("op_2726_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2726_end_mask_0 = const()[name = string("op_2726_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2726_squeeze_mask_0 = const()[name = string("op_2726_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2726_cast_fp16 = slice_by_index(begin = var_2726_begin_0, end = var_2726_end_0, end_mask = var_2726_end_mask_0, squeeze_mask = var_2726_squeeze_mask_0, x = var_2723_cast_fp16)[name = string("op_2726_cast_fp16")]; tensor var_2741_begin_0 = const()[name = string("op_2741_begin_0"), val = tensor([0, 2, 0, 0])]; tensor var_2741_end_0 = const()[name = string("op_2741_end_0"), val = tensor([1, 3, 1, 1536])]; tensor var_2741_end_mask_0 = const()[name = string("op_2741_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2741_cast_fp16 = slice_by_index(begin = var_2741_begin_0, end = var_2741_end_0, end_mask = var_2741_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2741_cast_fp16")]; tensor var_2744_begin_0 = const()[name = string("op_2744_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2744_end_0 = const()[name = string("op_2744_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2744_end_mask_0 = const()[name = string("op_2744_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2744_squeeze_mask_0 = const()[name = string("op_2744_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2744_cast_fp16 = slice_by_index(begin = var_2744_begin_0, end = var_2744_end_0, end_mask = var_2744_end_mask_0, squeeze_mask = var_2744_squeeze_mask_0, x = var_2741_cast_fp16)[name = string("op_2744_cast_fp16")]; tensor var_2759_begin_0 = const()[name = string("op_2759_begin_0"), val = tensor([0, 5, 0, 0])]; tensor var_2759_end_0 = const()[name = string("op_2759_end_0"), val = tensor([1, 6, 1, 1536])]; tensor var_2759_end_mask_0 = const()[name = string("op_2759_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2759_cast_fp16 = slice_by_index(begin = var_2759_begin_0, end = var_2759_end_0, end_mask = var_2759_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2759_cast_fp16")]; tensor var_2762_begin_0 = const()[name = string("op_2762_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2762_end_0 = const()[name = string("op_2762_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2762_end_mask_0 = const()[name = string("op_2762_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2762_squeeze_mask_0 = const()[name = string("op_2762_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2762_cast_fp16 = slice_by_index(begin = var_2762_begin_0, end = var_2762_end_0, end_mask = var_2762_end_mask_0, squeeze_mask = var_2762_squeeze_mask_0, x = var_2759_cast_fp16)[name = string("op_2762_cast_fp16")]; tensor var_2777_begin_0 = const()[name = string("op_2777_begin_0"), val = tensor([0, 7, 0, 0])]; tensor var_2777_end_0 = const()[name = string("op_2777_end_0"), val = tensor([1, 8, 1, 1536])]; tensor var_2777_end_mask_0 = const()[name = string("op_2777_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2777_cast_fp16 = slice_by_index(begin = var_2777_begin_0, end = var_2777_end_0, end_mask = var_2777_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2777_cast_fp16")]; tensor var_2780_begin_0 = const()[name = string("op_2780_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2780_end_0 = const()[name = string("op_2780_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2780_end_mask_0 = const()[name = string("op_2780_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2780_squeeze_mask_0 = const()[name = string("op_2780_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2780_cast_fp16 = slice_by_index(begin = var_2780_begin_0, end = var_2780_end_0, end_mask = var_2780_end_mask_0, squeeze_mask = var_2780_squeeze_mask_0, x = var_2777_cast_fp16)[name = string("op_2780_cast_fp16")]; tensor var_2795_begin_0 = const()[name = string("op_2795_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2795_end_0 = const()[name = string("op_2795_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2795_end_mask_0 = const()[name = string("op_2795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2795_cast_fp16 = slice_by_index(begin = var_2795_begin_0, end = var_2795_end_0, end_mask = var_2795_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2795_cast_fp16")]; tensor var_2798_begin_0 = const()[name = string("op_2798_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2798_end_0 = const()[name = string("op_2798_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2798_end_mask_0 = const()[name = string("op_2798_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2798_squeeze_mask_0 = const()[name = string("op_2798_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2798_cast_fp16 = slice_by_index(begin = var_2798_begin_0, end = var_2798_end_0, end_mask = var_2798_end_mask_0, squeeze_mask = var_2798_squeeze_mask_0, x = var_2795_cast_fp16)[name = string("op_2798_cast_fp16")]; tensor var_2813_begin_0 = const()[name = string("op_2813_begin_0"), val = tensor([0, 4, 0, 0])]; tensor var_2813_end_0 = const()[name = string("op_2813_end_0"), val = tensor([1, 5, 1, 1536])]; tensor var_2813_end_mask_0 = const()[name = string("op_2813_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2813_cast_fp16 = slice_by_index(begin = var_2813_begin_0, end = var_2813_end_0, end_mask = var_2813_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2813_cast_fp16")]; tensor var_2816_begin_0 = const()[name = string("op_2816_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2816_end_0 = const()[name = string("op_2816_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2816_end_mask_0 = const()[name = string("op_2816_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2816_squeeze_mask_0 = const()[name = string("op_2816_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2816_cast_fp16 = slice_by_index(begin = var_2816_begin_0, end = var_2816_end_0, end_mask = var_2816_end_mask_0, squeeze_mask = var_2816_squeeze_mask_0, x = var_2813_cast_fp16)[name = string("op_2816_cast_fp16")]; tensor var_2831_begin_0 = const()[name = string("op_2831_begin_0"), val = tensor([0, 8, 0, 0])]; tensor var_2831_end_0 = const()[name = string("op_2831_end_0"), val = tensor([1, 9, 1, 1536])]; tensor var_2831_end_mask_0 = const()[name = string("op_2831_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2831_cast_fp16 = slice_by_index(begin = var_2831_begin_0, end = var_2831_end_0, end_mask = var_2831_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2831_cast_fp16")]; tensor var_2834_begin_0 = const()[name = string("op_2834_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2834_end_0 = const()[name = string("op_2834_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2834_end_mask_0 = const()[name = string("op_2834_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2834_squeeze_mask_0 = const()[name = string("op_2834_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2834_cast_fp16 = slice_by_index(begin = var_2834_begin_0, end = var_2834_end_0, end_mask = var_2834_end_mask_0, squeeze_mask = var_2834_squeeze_mask_0, x = var_2831_cast_fp16)[name = string("op_2834_cast_fp16")]; tensor var_2849_begin_0 = const()[name = string("op_2849_begin_0"), val = tensor([0, 10, 0, 0])]; tensor var_2849_end_0 = const()[name = string("op_2849_end_0"), val = tensor([1, 11, 1, 1536])]; tensor var_2849_end_mask_0 = const()[name = string("op_2849_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2849_cast_fp16 = slice_by_index(begin = var_2849_begin_0, end = var_2849_end_0, end_mask = var_2849_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2849_cast_fp16")]; tensor var_2852_begin_0 = const()[name = string("op_2852_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2852_end_0 = const()[name = string("op_2852_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2852_end_mask_0 = const()[name = string("op_2852_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2852_squeeze_mask_0 = const()[name = string("op_2852_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2852_cast_fp16 = slice_by_index(begin = var_2852_begin_0, end = var_2852_end_0, end_mask = var_2852_end_mask_0, squeeze_mask = var_2852_squeeze_mask_0, x = var_2849_cast_fp16)[name = string("op_2852_cast_fp16")]; tensor var_2867_begin_0 = const()[name = string("op_2867_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2867_end_0 = const()[name = string("op_2867_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2867_end_mask_0 = const()[name = string("op_2867_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2867_cast_fp16 = slice_by_index(begin = var_2867_begin_0, end = var_2867_end_0, end_mask = var_2867_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2867_cast_fp16")]; tensor var_2870_begin_0 = const()[name = string("op_2870_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2870_end_0 = const()[name = string("op_2870_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2870_end_mask_0 = const()[name = string("op_2870_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2870_squeeze_mask_0 = const()[name = string("op_2870_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2870_cast_fp16 = slice_by_index(begin = var_2870_begin_0, end = var_2870_end_0, end_mask = var_2870_end_mask_0, squeeze_mask = var_2870_squeeze_mask_0, x = var_2867_cast_fp16)[name = string("op_2870_cast_fp16")]; tensor var_2885_begin_0 = const()[name = string("op_2885_begin_0"), val = tensor([0, 1, 0, 0])]; tensor var_2885_end_0 = const()[name = string("op_2885_end_0"), val = tensor([1, 2, 1, 1536])]; tensor var_2885_end_mask_0 = const()[name = string("op_2885_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2885_cast_fp16 = slice_by_index(begin = var_2885_begin_0, end = var_2885_end_0, end_mask = var_2885_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2885_cast_fp16")]; tensor var_2888_begin_0 = const()[name = string("op_2888_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2888_end_0 = const()[name = string("op_2888_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2888_end_mask_0 = const()[name = string("op_2888_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2888_squeeze_mask_0 = const()[name = string("op_2888_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2888_cast_fp16 = slice_by_index(begin = var_2888_begin_0, end = var_2888_end_0, end_mask = var_2888_end_mask_0, squeeze_mask = var_2888_squeeze_mask_0, x = var_2885_cast_fp16)[name = string("op_2888_cast_fp16")]; tensor var_2903_begin_0 = const()[name = string("op_2903_begin_0"), val = tensor([0, 2, 0, 0])]; tensor var_2903_end_0 = const()[name = string("op_2903_end_0"), val = tensor([1, 3, 1, 1536])]; tensor var_2903_end_mask_0 = const()[name = string("op_2903_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2903_cast_fp16 = slice_by_index(begin = var_2903_begin_0, end = var_2903_end_0, end_mask = var_2903_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2903_cast_fp16")]; tensor var_2906_begin_0 = const()[name = string("op_2906_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2906_end_0 = const()[name = string("op_2906_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2906_end_mask_0 = const()[name = string("op_2906_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2906_squeeze_mask_0 = const()[name = string("op_2906_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2906_cast_fp16 = slice_by_index(begin = var_2906_begin_0, end = var_2906_end_0, end_mask = var_2906_end_mask_0, squeeze_mask = var_2906_squeeze_mask_0, x = var_2903_cast_fp16)[name = string("op_2906_cast_fp16")]; tensor var_2921_begin_0 = const()[name = string("op_2921_begin_0"), val = tensor([0, 3, 0, 0])]; tensor var_2921_end_0 = const()[name = string("op_2921_end_0"), val = tensor([1, 4, 1, 1536])]; tensor var_2921_end_mask_0 = const()[name = string("op_2921_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2921_cast_fp16 = slice_by_index(begin = var_2921_begin_0, end = var_2921_end_0, end_mask = var_2921_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2921_cast_fp16")]; tensor var_2924_begin_0 = const()[name = string("op_2924_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2924_end_0 = const()[name = string("op_2924_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2924_end_mask_0 = const()[name = string("op_2924_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2924_squeeze_mask_0 = const()[name = string("op_2924_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2924_cast_fp16 = slice_by_index(begin = var_2924_begin_0, end = var_2924_end_0, end_mask = var_2924_end_mask_0, squeeze_mask = var_2924_squeeze_mask_0, x = var_2921_cast_fp16)[name = string("op_2924_cast_fp16")]; tensor var_2939_begin_0 = const()[name = string("op_2939_begin_0"), val = tensor([0, 6, 0, 0])]; tensor var_2939_end_0 = const()[name = string("op_2939_end_0"), val = tensor([1, 7, 1, 1536])]; tensor var_2939_end_mask_0 = const()[name = string("op_2939_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2939_cast_fp16 = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2939_cast_fp16")]; tensor var_2942_begin_0 = const()[name = string("op_2942_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2942_end_0 = const()[name = string("op_2942_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2942_end_mask_0 = const()[name = string("op_2942_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2942_squeeze_mask_0 = const()[name = string("op_2942_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2942_cast_fp16 = slice_by_index(begin = var_2942_begin_0, end = var_2942_end_0, end_mask = var_2942_end_mask_0, squeeze_mask = var_2942_squeeze_mask_0, x = var_2939_cast_fp16)[name = string("op_2942_cast_fp16")]; tensor var_2957_begin_0 = const()[name = string("op_2957_begin_0"), val = tensor([0, 11, 0, 0])]; tensor var_2957_end_0 = const()[name = string("op_2957_end_0"), val = tensor([1, 12, 1, 1536])]; tensor var_2957_end_mask_0 = const()[name = string("op_2957_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2957_cast_fp16 = slice_by_index(begin = var_2957_begin_0, end = var_2957_end_0, end_mask = var_2957_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2957_cast_fp16")]; tensor var_2960_begin_0 = const()[name = string("op_2960_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2960_end_0 = const()[name = string("op_2960_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2960_end_mask_0 = const()[name = string("op_2960_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2960_squeeze_mask_0 = const()[name = string("op_2960_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2960_cast_fp16 = slice_by_index(begin = var_2960_begin_0, end = var_2960_end_0, end_mask = var_2960_end_mask_0, squeeze_mask = var_2960_squeeze_mask_0, x = var_2957_cast_fp16)[name = string("op_2960_cast_fp16")]; tensor var_2975_begin_0 = const()[name = string("op_2975_begin_0"), val = tensor([0, 2, 0, 0])]; tensor var_2975_end_0 = const()[name = string("op_2975_end_0"), val = tensor([1, 3, 1, 1536])]; tensor var_2975_end_mask_0 = const()[name = string("op_2975_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2975_cast_fp16 = slice_by_index(begin = var_2975_begin_0, end = var_2975_end_0, end_mask = var_2975_end_mask_0, x = obj_221_cast_fp16)[name = string("op_2975_cast_fp16")]; tensor var_2978_begin_0 = const()[name = string("op_2978_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2978_end_0 = const()[name = string("op_2978_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2978_end_mask_0 = const()[name = string("op_2978_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2978_squeeze_mask_0 = const()[name = string("op_2978_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2978_cast_fp16 = slice_by_index(begin = var_2978_begin_0, end = var_2978_end_0, end_mask = var_2978_end_mask_0, squeeze_mask = var_2978_squeeze_mask_0, x = var_2975_cast_fp16)[name = string("op_2978_cast_fp16")]; tensor var_2993_begin_0 = const()[name = string("op_2993_begin_0"), val = tensor([0, 4, 0, 0])]; tensor var_2993_end_0 = const()[name = string("op_2993_end_0"), val = tensor([1, 5, 1, 1536])]; tensor var_2993_end_mask_0 = const()[name = string("op_2993_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2993_cast_fp16 = slice_by_index(begin = var_2993_begin_0, end = var_2993_end_0, end_mask = var_2993_end_mask_0, x = obj_221_cast_fp16)[name = string("op_2993_cast_fp16")]; tensor var_2996_begin_0 = const()[name = string("op_2996_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2996_end_0 = const()[name = string("op_2996_end_0"), val = tensor([1, 1, 1, 1536])]; tensor var_2996_end_mask_0 = const()[name = string("op_2996_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_2996_squeeze_mask_0 = const()[name = string("op_2996_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_2996_cast_fp16 = slice_by_index(begin = var_2996_begin_0, end = var_2996_end_0, end_mask = var_2996_end_mask_0, squeeze_mask = var_2996_squeeze_mask_0, x = var_2993_cast_fp16)[name = string("op_2996_cast_fp16")]; int32 var_3003 = const()[name = string("op_3003"), val = int32(1)]; bool var_3004_interleave_0 = const()[name = string("op_3004_interleave_0"), val = bool(false)]; tensor var_3004_cast_fp16 = concat(axis = var_3003, interleave = var_3004_interleave_0, values = (var_2672_cast_fp16, var_2690_cast_fp16, var_2708_cast_fp16, var_2726_cast_fp16, var_2744_cast_fp16, var_2762_cast_fp16, var_2780_cast_fp16, var_2798_cast_fp16, var_2816_cast_fp16, var_2834_cast_fp16, var_2852_cast_fp16, var_2870_cast_fp16, var_2888_cast_fp16, var_2906_cast_fp16, var_2924_cast_fp16, var_2942_cast_fp16, var_2960_cast_fp16, var_2978_cast_fp16, var_2996_cast_fp16))[name = string("op_3004_cast_fp16")]; bool var_3007 = const()[name = string("op_3007"), val = bool(false)]; tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_3007, x = var_3004_cast_fp16)[name = string("obj_cast_fp16")]; } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); }