| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.4"}, {"coremlc-version", "1436.100.10"}})] | |
| { | |
| func main<ios16>(tensor<int32, [1, 77]> input_ids) { | |
| tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, []> inputs_embeds_axis_0 = const()[name = tensor<string, []>("inputs_embeds_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [49408, 768]> text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [49408, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [1, 77, 768]> inputs_embeds_cast = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = input_ids, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast")]; | |
| tensor<fp16, [1, 77, 768]> position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [44352]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935232))), name = tensor<string, []>("position_embeddings_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 768]> input_3_cast = add(x = inputs_embeds_cast, y = position_embeddings_to_fp16_palettized)[name = tensor<string, []>("input_3_cast")]; | |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935424)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75937024)))]; | |
| tensor<fp16, []> var_13_to_fp16 = const()[name = tensor<string, []>("op_13_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_1_cast = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast)[name = tensor<string, []>("hidden_states_1_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75938624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381248)))]; | |
| tensor<fp16, [1, 77, 768]> var_87_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast)[name = tensor<string, []>("op_87_cast")]; | |
| tensor<fp16, []> var_88_to_fp16 = const()[name = tensor<string, []>("op_88_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_5_cast = mul(x = var_87_cast, y = var_88_to_fp16)[name = tensor<string, []>("tensor_5_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76382848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825472)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_1_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast)[name = tensor<string, []>("tensor_1_cast")]; | |
| tensor<int32, [4]> var_93 = const()[name = tensor<string, []>("op_93"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_94_cast = reshape(shape = var_93, x = tensor_1_cast)[name = tensor<string, []>("op_94_cast")]; | |
| tensor<int32, [4]> var_95_perm_0 = const()[name = tensor<string, []>("op_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76827072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269696)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_3_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast)[name = tensor<string, []>("tensor_3_cast")]; | |
| tensor<int32, [4]> var_100 = const()[name = tensor<string, []>("op_100"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_101_cast = reshape(shape = var_100, x = tensor_3_cast)[name = tensor<string, []>("op_101_cast")]; | |
| tensor<int32, [4]> var_102_perm_0 = const()[name = tensor<string, []>("op_102_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_109 = const()[name = tensor<string, []>("op_109"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_110_cast = reshape(shape = var_109, x = tensor_5_cast)[name = tensor<string, []>("op_110_cast")]; | |
| tensor<int32, [4]> var_111_perm_0 = const()[name = tensor<string, []>("op_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_113 = const()[name = tensor<string, []>("op_113"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_59 = transpose(perm = var_111_perm_0, x = var_110_cast)[name = tensor<string, []>("transpose_59")]; | |
| tensor<fp16, [12, 77, 64]> query_states_1_cast = reshape(shape = var_113, x = transpose_59)[name = tensor<string, []>("query_states_1_cast")]; | |
| tensor<int32, [3]> var_115 = const()[name = tensor<string, []>("op_115"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_58 = transpose(perm = var_95_perm_0, x = var_94_cast)[name = tensor<string, []>("transpose_58")]; | |
| tensor<fp16, [12, 77, 64]> key_states_3_cast = reshape(shape = var_115, x = transpose_58)[name = tensor<string, []>("key_states_3_cast")]; | |
| tensor<int32, [3]> var_117 = const()[name = tensor<string, []>("op_117"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_57 = transpose(perm = var_102_perm_0, x = var_101_cast)[name = tensor<string, []>("transpose_57")]; | |
| tensor<fp16, [12, 77, 64]> value_states_3_cast = reshape(shape = var_117, x = transpose_57)[name = tensor<string, []>("value_states_3_cast")]; | |
| tensor<int32, [3]> var_120_perm_0 = const()[name = tensor<string, []>("op_120_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_1_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_1_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_56 = transpose(perm = var_120_perm_0, x = key_states_3_cast)[name = tensor<string, []>("transpose_56")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_1_cast = matmul(transpose_x = attn_weights_1_transpose_x_0, transpose_y = attn_weights_1_transpose_y_0, x = query_states_1_cast, y = transpose_56)[name = tensor<string, []>("attn_weights_1_cast")]; | |
| tensor<int32, [4]> var_122 = const()[name = tensor<string, []>("op_122"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_123_cast = reshape(shape = var_122, x = attn_weights_1_cast)[name = tensor<string, []>("op_123_cast")]; | |
| tensor<fp16, [1, 1, 77, 77]> causal_attention_mask_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4447]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77271296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77275840))), name = tensor<string, []>("causal_attention_mask_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_3_cast = add(x = var_123_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_3_cast")]; | |
| tensor<int32, [3]> var_128 = const()[name = tensor<string, []>("op_128"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_5_cast = reshape(shape = var_128, x = attn_weights_3_cast)[name = tensor<string, []>("input_5_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_7_cast = softmax(axis = var_5, x = input_5_cast)[name = tensor<string, []>("input_7_cast")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_1_cast = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast, y = value_states_3_cast)[name = tensor<string, []>("attn_output_1_cast")]; | |
| tensor<int32, [4]> var_133 = const()[name = tensor<string, []>("op_133"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_3_cast = reshape(shape = var_133, x = attn_output_1_cast)[name = tensor<string, []>("attn_output_3_cast")]; | |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_136 = const()[name = tensor<string, []>("op_136"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_55 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast)[name = tensor<string, []>("transpose_55")]; | |
| tensor<fp16, [1, 77, 768]> input_9_cast = reshape(shape = var_136, x = transpose_55)[name = tensor<string, []>("input_9_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77276032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718656)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_3_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast)[name = tensor<string, []>("hidden_states_3_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_11_cast = add(x = input_3_cast, y = hidden_states_3_cast)[name = tensor<string, []>("input_11_cast")]; | |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77720256)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77721856)))]; | |
| tensor<fp16, [1, 77, 768]> input_13_cast = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast)[name = tensor<string, []>("input_13_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77723456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79492992))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79493184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_15_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast)[name = tensor<string, []>("input_15_cast")]; | |
| tensor<fp16, []> var_151_to_fp16 = const()[name = tensor<string, []>("op_151_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_152_cast = mul(x = input_15_cast, y = var_151_to_fp16)[name = tensor<string, []>("op_152_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_153_cast = sigmoid(x = var_152_cast)[name = tensor<string, []>("op_153_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_17_cast = mul(x = input_15_cast, y = var_153_cast)[name = tensor<string, []>("input_17_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81265280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81265472)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_5_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast)[name = tensor<string, []>("hidden_states_5_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_19_cast = add(x = input_11_cast, y = hidden_states_5_cast)[name = tensor<string, []>("input_19_cast")]; | |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81267072)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81268672)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_7_cast = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast)[name = tensor<string, []>("hidden_states_7_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81270272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81712704))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81712896)))]; | |
| tensor<fp16, [1, 77, 768]> var_177_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast)[name = tensor<string, []>("op_177_cast")]; | |
| tensor<fp16, []> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_11_cast = mul(x = var_177_cast, y = var_178_to_fp16)[name = tensor<string, []>("tensor_11_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81714496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82156928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82157120)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_7_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast)[name = tensor<string, []>("tensor_7_cast")]; | |
| tensor<int32, [4]> var_183 = const()[name = tensor<string, []>("op_183"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_184_cast = reshape(shape = var_183, x = tensor_7_cast)[name = tensor<string, []>("op_184_cast")]; | |
| tensor<int32, [4]> var_185_perm_0 = const()[name = tensor<string, []>("op_185_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82158720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82601152))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82601344)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_9_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast)[name = tensor<string, []>("tensor_9_cast")]; | |
| tensor<int32, [4]> var_190 = const()[name = tensor<string, []>("op_190"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_191_cast = reshape(shape = var_190, x = tensor_9_cast)[name = tensor<string, []>("op_191_cast")]; | |
| tensor<int32, [4]> var_192_perm_0 = const()[name = tensor<string, []>("op_192_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_199 = const()[name = tensor<string, []>("op_199"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_200_cast = reshape(shape = var_199, x = tensor_11_cast)[name = tensor<string, []>("op_200_cast")]; | |
| tensor<int32, [4]> var_201_perm_0 = const()[name = tensor<string, []>("op_201_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_203 = const()[name = tensor<string, []>("op_203"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_54 = transpose(perm = var_201_perm_0, x = var_200_cast)[name = tensor<string, []>("transpose_54")]; | |
| tensor<fp16, [12, 77, 64]> query_states_3_cast = reshape(shape = var_203, x = transpose_54)[name = tensor<string, []>("query_states_3_cast")]; | |
| tensor<int32, [3]> var_205 = const()[name = tensor<string, []>("op_205"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_53 = transpose(perm = var_185_perm_0, x = var_184_cast)[name = tensor<string, []>("transpose_53")]; | |
| tensor<fp16, [12, 77, 64]> key_states_7_cast = reshape(shape = var_205, x = transpose_53)[name = tensor<string, []>("key_states_7_cast")]; | |
| tensor<int32, [3]> var_207 = const()[name = tensor<string, []>("op_207"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_52 = transpose(perm = var_192_perm_0, x = var_191_cast)[name = tensor<string, []>("transpose_52")]; | |
| tensor<fp16, [12, 77, 64]> value_states_7_cast = reshape(shape = var_207, x = transpose_52)[name = tensor<string, []>("value_states_7_cast")]; | |
| tensor<int32, [3]> var_210_perm_0 = const()[name = tensor<string, []>("op_210_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_7_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_7_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_51 = transpose(perm = var_210_perm_0, x = key_states_7_cast)[name = tensor<string, []>("transpose_51")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_7_cast = matmul(transpose_x = attn_weights_7_transpose_x_0, transpose_y = attn_weights_7_transpose_y_0, x = query_states_3_cast, y = transpose_51)[name = tensor<string, []>("attn_weights_7_cast")]; | |
| tensor<int32, [4]> var_212 = const()[name = tensor<string, []>("op_212"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_213_cast = reshape(shape = var_212, x = attn_weights_7_cast)[name = tensor<string, []>("op_213_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_9_cast = add(x = var_213_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_9_cast")]; | |
| tensor<int32, [3]> var_218 = const()[name = tensor<string, []>("op_218"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_21_cast = reshape(shape = var_218, x = attn_weights_9_cast)[name = tensor<string, []>("input_21_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_23_cast = softmax(axis = var_5, x = input_21_cast)[name = tensor<string, []>("input_23_cast")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_7_cast = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast, y = value_states_7_cast)[name = tensor<string, []>("attn_output_7_cast")]; | |
| tensor<int32, [4]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast = reshape(shape = var_223, x = attn_output_7_cast)[name = tensor<string, []>("attn_output_9_cast")]; | |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_50 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast)[name = tensor<string, []>("transpose_50")]; | |
| tensor<fp16, [1, 77, 768]> input_25_cast = reshape(shape = var_226, x = transpose_50)[name = tensor<string, []>("input_25_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82602944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83045376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83045568)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_9_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast)[name = tensor<string, []>("hidden_states_9_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_27_cast = add(x = input_19_cast, y = hidden_states_9_cast)[name = tensor<string, []>("input_27_cast")]; | |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83047168)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83048768)))]; | |
| tensor<fp16, [1, 77, 768]> input_29_cast = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast)[name = tensor<string, []>("input_29_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83050368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84819904))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84820096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_31_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast)[name = tensor<string, []>("input_31_cast")]; | |
| tensor<fp16, []> var_241_to_fp16 = const()[name = tensor<string, []>("op_241_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_242_cast = mul(x = input_31_cast, y = var_241_to_fp16)[name = tensor<string, []>("op_242_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_243_cast = sigmoid(x = var_242_cast)[name = tensor<string, []>("op_243_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_33_cast = mul(x = input_31_cast, y = var_243_cast)[name = tensor<string, []>("input_33_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86592192))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86592384)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_11_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast)[name = tensor<string, []>("hidden_states_11_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_35_cast = add(x = input_27_cast, y = hidden_states_11_cast)[name = tensor<string, []>("input_35_cast")]; | |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86593984)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86595584)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_13_cast = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast)[name = tensor<string, []>("hidden_states_13_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86597184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87039616))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87039808)))]; | |
| tensor<fp16, [1, 77, 768]> var_267_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast)[name = tensor<string, []>("op_267_cast")]; | |
| tensor<fp16, []> var_268_to_fp16 = const()[name = tensor<string, []>("op_268_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_17_cast = mul(x = var_267_cast, y = var_268_to_fp16)[name = tensor<string, []>("tensor_17_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87041408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87483840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87484032)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_13_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast)[name = tensor<string, []>("tensor_13_cast")]; | |
| tensor<int32, [4]> var_273 = const()[name = tensor<string, []>("op_273"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_274_cast = reshape(shape = var_273, x = tensor_13_cast)[name = tensor<string, []>("op_274_cast")]; | |
| tensor<int32, [4]> var_275_perm_0 = const()[name = tensor<string, []>("op_275_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87485632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87928064))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87928256)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_15_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast)[name = tensor<string, []>("tensor_15_cast")]; | |
| tensor<int32, [4]> var_280 = const()[name = tensor<string, []>("op_280"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_281_cast = reshape(shape = var_280, x = tensor_15_cast)[name = tensor<string, []>("op_281_cast")]; | |
| tensor<int32, [4]> var_282_perm_0 = const()[name = tensor<string, []>("op_282_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_289 = const()[name = tensor<string, []>("op_289"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_290_cast = reshape(shape = var_289, x = tensor_17_cast)[name = tensor<string, []>("op_290_cast")]; | |
| tensor<int32, [4]> var_291_perm_0 = const()[name = tensor<string, []>("op_291_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_293 = const()[name = tensor<string, []>("op_293"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_49 = transpose(perm = var_291_perm_0, x = var_290_cast)[name = tensor<string, []>("transpose_49")]; | |
| tensor<fp16, [12, 77, 64]> query_states_5_cast = reshape(shape = var_293, x = transpose_49)[name = tensor<string, []>("query_states_5_cast")]; | |
| tensor<int32, [3]> var_295 = const()[name = tensor<string, []>("op_295"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_48 = transpose(perm = var_275_perm_0, x = var_274_cast)[name = tensor<string, []>("transpose_48")]; | |
| tensor<fp16, [12, 77, 64]> key_states_11_cast = reshape(shape = var_295, x = transpose_48)[name = tensor<string, []>("key_states_11_cast")]; | |
| tensor<int32, [3]> var_297 = const()[name = tensor<string, []>("op_297"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_47 = transpose(perm = var_282_perm_0, x = var_281_cast)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [12, 77, 64]> value_states_11_cast = reshape(shape = var_297, x = transpose_47)[name = tensor<string, []>("value_states_11_cast")]; | |
| tensor<int32, [3]> var_300_perm_0 = const()[name = tensor<string, []>("op_300_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_13_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_13_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_46 = transpose(perm = var_300_perm_0, x = key_states_11_cast)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_13_cast = matmul(transpose_x = attn_weights_13_transpose_x_0, transpose_y = attn_weights_13_transpose_y_0, x = query_states_5_cast, y = transpose_46)[name = tensor<string, []>("attn_weights_13_cast")]; | |
| tensor<int32, [4]> var_302 = const()[name = tensor<string, []>("op_302"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_303_cast = reshape(shape = var_302, x = attn_weights_13_cast)[name = tensor<string, []>("op_303_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_15_cast = add(x = var_303_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_15_cast")]; | |
| tensor<int32, [3]> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_37_cast = reshape(shape = var_308, x = attn_weights_15_cast)[name = tensor<string, []>("input_37_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_39_cast = softmax(axis = var_5, x = input_37_cast)[name = tensor<string, []>("input_39_cast")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_13_cast = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast, y = value_states_11_cast)[name = tensor<string, []>("attn_output_13_cast")]; | |
| tensor<int32, [4]> var_313 = const()[name = tensor<string, []>("op_313"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_15_cast = reshape(shape = var_313, x = attn_output_13_cast)[name = tensor<string, []>("attn_output_15_cast")]; | |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_316 = const()[name = tensor<string, []>("op_316"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_45 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [1, 77, 768]> input_41_cast = reshape(shape = var_316, x = transpose_45)[name = tensor<string, []>("input_41_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87929856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88372288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88372480)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_15_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast)[name = tensor<string, []>("hidden_states_15_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_43_cast = add(x = input_35_cast, y = hidden_states_15_cast)[name = tensor<string, []>("input_43_cast")]; | |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88374080)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88375680)))]; | |
| tensor<fp16, [1, 77, 768]> input_45_cast = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast)[name = tensor<string, []>("input_45_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88377280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90146816))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90147008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_47_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast)[name = tensor<string, []>("input_47_cast")]; | |
| tensor<fp16, []> var_331_to_fp16 = const()[name = tensor<string, []>("op_331_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_332_cast = mul(x = input_47_cast, y = var_331_to_fp16)[name = tensor<string, []>("op_332_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_333_cast = sigmoid(x = var_332_cast)[name = tensor<string, []>("op_333_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_49_cast = mul(x = input_47_cast, y = var_333_cast)[name = tensor<string, []>("input_49_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91919104))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91919296)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_17_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast)[name = tensor<string, []>("hidden_states_17_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_51_cast = add(x = input_43_cast, y = hidden_states_17_cast)[name = tensor<string, []>("input_51_cast")]; | |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91920896)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91922496)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_19_cast = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast)[name = tensor<string, []>("hidden_states_19_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91924096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366528))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366720)))]; | |
| tensor<fp16, [1, 77, 768]> var_357_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast)[name = tensor<string, []>("op_357_cast")]; | |
| tensor<fp16, []> var_358_to_fp16 = const()[name = tensor<string, []>("op_358_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_23_cast = mul(x = var_357_cast, y = var_358_to_fp16)[name = tensor<string, []>("tensor_23_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92368320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92810752))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92810944)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_19_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast)[name = tensor<string, []>("tensor_19_cast")]; | |
| tensor<int32, [4]> var_363 = const()[name = tensor<string, []>("op_363"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_364_cast = reshape(shape = var_363, x = tensor_19_cast)[name = tensor<string, []>("op_364_cast")]; | |
| tensor<int32, [4]> var_365_perm_0 = const()[name = tensor<string, []>("op_365_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92812544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93254976))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93255168)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_21_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast)[name = tensor<string, []>("tensor_21_cast")]; | |
| tensor<int32, [4]> var_370 = const()[name = tensor<string, []>("op_370"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_371_cast = reshape(shape = var_370, x = tensor_21_cast)[name = tensor<string, []>("op_371_cast")]; | |
| tensor<int32, [4]> var_372_perm_0 = const()[name = tensor<string, []>("op_372_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_379 = const()[name = tensor<string, []>("op_379"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_380_cast = reshape(shape = var_379, x = tensor_23_cast)[name = tensor<string, []>("op_380_cast")]; | |
| tensor<int32, [4]> var_381_perm_0 = const()[name = tensor<string, []>("op_381_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_383 = const()[name = tensor<string, []>("op_383"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_44 = transpose(perm = var_381_perm_0, x = var_380_cast)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [12, 77, 64]> query_states_7_cast = reshape(shape = var_383, x = transpose_44)[name = tensor<string, []>("query_states_7_cast")]; | |
| tensor<int32, [3]> var_385 = const()[name = tensor<string, []>("op_385"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_43 = transpose(perm = var_365_perm_0, x = var_364_cast)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [12, 77, 64]> key_states_15_cast = reshape(shape = var_385, x = transpose_43)[name = tensor<string, []>("key_states_15_cast")]; | |
| tensor<int32, [3]> var_387 = const()[name = tensor<string, []>("op_387"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_42 = transpose(perm = var_372_perm_0, x = var_371_cast)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [12, 77, 64]> value_states_15_cast = reshape(shape = var_387, x = transpose_42)[name = tensor<string, []>("value_states_15_cast")]; | |
| tensor<int32, [3]> var_390_perm_0 = const()[name = tensor<string, []>("op_390_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_19_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_19_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_41 = transpose(perm = var_390_perm_0, x = key_states_15_cast)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_19_cast = matmul(transpose_x = attn_weights_19_transpose_x_0, transpose_y = attn_weights_19_transpose_y_0, x = query_states_7_cast, y = transpose_41)[name = tensor<string, []>("attn_weights_19_cast")]; | |
| tensor<int32, [4]> var_392 = const()[name = tensor<string, []>("op_392"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_393_cast = reshape(shape = var_392, x = attn_weights_19_cast)[name = tensor<string, []>("op_393_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_21_cast = add(x = var_393_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_21_cast")]; | |
| tensor<int32, [3]> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_53_cast = reshape(shape = var_398, x = attn_weights_21_cast)[name = tensor<string, []>("input_53_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_55_cast = softmax(axis = var_5, x = input_53_cast)[name = tensor<string, []>("input_55_cast")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_19_cast = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast, y = value_states_15_cast)[name = tensor<string, []>("attn_output_19_cast")]; | |
| tensor<int32, [4]> var_403 = const()[name = tensor<string, []>("op_403"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast = reshape(shape = var_403, x = attn_output_19_cast)[name = tensor<string, []>("attn_output_21_cast")]; | |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_406 = const()[name = tensor<string, []>("op_406"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_40 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [1, 77, 768]> input_57_cast = reshape(shape = var_406, x = transpose_40)[name = tensor<string, []>("input_57_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93256768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93699200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93699392)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_21_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast)[name = tensor<string, []>("hidden_states_21_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_59_cast = add(x = input_51_cast, y = hidden_states_21_cast)[name = tensor<string, []>("input_59_cast")]; | |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93700992)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93702592)))]; | |
| tensor<fp16, [1, 77, 768]> input_61_cast = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast)[name = tensor<string, []>("input_61_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93704192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473728))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_63_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast)[name = tensor<string, []>("input_63_cast")]; | |
| tensor<fp16, []> var_421_to_fp16 = const()[name = tensor<string, []>("op_421_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_422_cast = mul(x = input_63_cast, y = var_421_to_fp16)[name = tensor<string, []>("op_422_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_423_cast = sigmoid(x = var_422_cast)[name = tensor<string, []>("op_423_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_65_cast = mul(x = input_63_cast, y = var_423_cast)[name = tensor<string, []>("input_65_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97246016))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97246208)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_23_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast)[name = tensor<string, []>("hidden_states_23_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_67_cast = add(x = input_59_cast, y = hidden_states_23_cast)[name = tensor<string, []>("input_67_cast")]; | |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97247808)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97249408)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_25_cast = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast)[name = tensor<string, []>("hidden_states_25_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97251008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97693440))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97693632)))]; | |
| tensor<fp16, [1, 77, 768]> var_447_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast)[name = tensor<string, []>("op_447_cast")]; | |
| tensor<fp16, []> var_448_to_fp16 = const()[name = tensor<string, []>("op_448_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_29_cast = mul(x = var_447_cast, y = var_448_to_fp16)[name = tensor<string, []>("tensor_29_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97695232))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98137664))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98137856)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_25_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast)[name = tensor<string, []>("tensor_25_cast")]; | |
| tensor<int32, [4]> var_453 = const()[name = tensor<string, []>("op_453"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_454_cast = reshape(shape = var_453, x = tensor_25_cast)[name = tensor<string, []>("op_454_cast")]; | |
| tensor<int32, [4]> var_455_perm_0 = const()[name = tensor<string, []>("op_455_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98139456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98581888))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98582080)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_27_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast)[name = tensor<string, []>("tensor_27_cast")]; | |
| tensor<int32, [4]> var_460 = const()[name = tensor<string, []>("op_460"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_461_cast = reshape(shape = var_460, x = tensor_27_cast)[name = tensor<string, []>("op_461_cast")]; | |
| tensor<int32, [4]> var_462_perm_0 = const()[name = tensor<string, []>("op_462_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_469 = const()[name = tensor<string, []>("op_469"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_470_cast = reshape(shape = var_469, x = tensor_29_cast)[name = tensor<string, []>("op_470_cast")]; | |
| tensor<int32, [4]> var_471_perm_0 = const()[name = tensor<string, []>("op_471_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_473 = const()[name = tensor<string, []>("op_473"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_39 = transpose(perm = var_471_perm_0, x = var_470_cast)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [12, 77, 64]> query_states_9_cast = reshape(shape = var_473, x = transpose_39)[name = tensor<string, []>("query_states_9_cast")]; | |
| tensor<int32, [3]> var_475 = const()[name = tensor<string, []>("op_475"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_38 = transpose(perm = var_455_perm_0, x = var_454_cast)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [12, 77, 64]> key_states_19_cast = reshape(shape = var_475, x = transpose_38)[name = tensor<string, []>("key_states_19_cast")]; | |
| tensor<int32, [3]> var_477 = const()[name = tensor<string, []>("op_477"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_37 = transpose(perm = var_462_perm_0, x = var_461_cast)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [12, 77, 64]> value_states_19_cast = reshape(shape = var_477, x = transpose_37)[name = tensor<string, []>("value_states_19_cast")]; | |
| tensor<int32, [3]> var_480_perm_0 = const()[name = tensor<string, []>("op_480_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_25_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_25_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_36 = transpose(perm = var_480_perm_0, x = key_states_19_cast)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_25_cast = matmul(transpose_x = attn_weights_25_transpose_x_0, transpose_y = attn_weights_25_transpose_y_0, x = query_states_9_cast, y = transpose_36)[name = tensor<string, []>("attn_weights_25_cast")]; | |
| tensor<int32, [4]> var_482 = const()[name = tensor<string, []>("op_482"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_483_cast = reshape(shape = var_482, x = attn_weights_25_cast)[name = tensor<string, []>("op_483_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_27_cast = add(x = var_483_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_27_cast")]; | |
| tensor<int32, [3]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_69_cast = reshape(shape = var_488, x = attn_weights_27_cast)[name = tensor<string, []>("input_69_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_71_cast = softmax(axis = var_5, x = input_69_cast)[name = tensor<string, []>("input_71_cast")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_25_cast = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast, y = value_states_19_cast)[name = tensor<string, []>("attn_output_25_cast")]; | |
| tensor<int32, [4]> var_493 = const()[name = tensor<string, []>("op_493"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_27_cast = reshape(shape = var_493, x = attn_output_25_cast)[name = tensor<string, []>("attn_output_27_cast")]; | |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_496 = const()[name = tensor<string, []>("op_496"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_35 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [1, 77, 768]> input_73_cast = reshape(shape = var_496, x = transpose_35)[name = tensor<string, []>("input_73_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98583680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99026112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99026304)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_27_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast)[name = tensor<string, []>("hidden_states_27_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_75_cast = add(x = input_67_cast, y = hidden_states_27_cast)[name = tensor<string, []>("input_75_cast")]; | |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99027904)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99029504)))]; | |
| tensor<fp16, [1, 77, 768]> input_77_cast = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast)[name = tensor<string, []>("input_77_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99031104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800640))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_79_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast)[name = tensor<string, []>("input_79_cast")]; | |
| tensor<fp16, []> var_511_to_fp16 = const()[name = tensor<string, []>("op_511_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_512_cast = mul(x = input_79_cast, y = var_511_to_fp16)[name = tensor<string, []>("op_512_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_513_cast = sigmoid(x = var_512_cast)[name = tensor<string, []>("op_513_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_81_cast = mul(x = input_79_cast, y = var_513_cast)[name = tensor<string, []>("input_81_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102572928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102573120)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_29_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast)[name = tensor<string, []>("hidden_states_29_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_83_cast = add(x = input_75_cast, y = hidden_states_29_cast)[name = tensor<string, []>("input_83_cast")]; | |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102574720)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102576320)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_31_cast = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast)[name = tensor<string, []>("hidden_states_31_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102577920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103020352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103020544)))]; | |
| tensor<fp16, [1, 77, 768]> var_537_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast)[name = tensor<string, []>("op_537_cast")]; | |
| tensor<fp16, []> var_538_to_fp16 = const()[name = tensor<string, []>("op_538_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_35_cast = mul(x = var_537_cast, y = var_538_to_fp16)[name = tensor<string, []>("tensor_35_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103022144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103464576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103464768)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_31_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast)[name = tensor<string, []>("tensor_31_cast")]; | |
| tensor<int32, [4]> var_543 = const()[name = tensor<string, []>("op_543"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_544_cast = reshape(shape = var_543, x = tensor_31_cast)[name = tensor<string, []>("op_544_cast")]; | |
| tensor<int32, [4]> var_545_perm_0 = const()[name = tensor<string, []>("op_545_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103466368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103908800))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103908992)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_33_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast)[name = tensor<string, []>("tensor_33_cast")]; | |
| tensor<int32, [4]> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_551_cast = reshape(shape = var_550, x = tensor_33_cast)[name = tensor<string, []>("op_551_cast")]; | |
| tensor<int32, [4]> var_552_perm_0 = const()[name = tensor<string, []>("op_552_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_559 = const()[name = tensor<string, []>("op_559"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_560_cast = reshape(shape = var_559, x = tensor_35_cast)[name = tensor<string, []>("op_560_cast")]; | |
| tensor<int32, [4]> var_561_perm_0 = const()[name = tensor<string, []>("op_561_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_563 = const()[name = tensor<string, []>("op_563"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_34 = transpose(perm = var_561_perm_0, x = var_560_cast)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [12, 77, 64]> query_states_11_cast = reshape(shape = var_563, x = transpose_34)[name = tensor<string, []>("query_states_11_cast")]; | |
| tensor<int32, [3]> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_33 = transpose(perm = var_545_perm_0, x = var_544_cast)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [12, 77, 64]> key_states_23_cast = reshape(shape = var_565, x = transpose_33)[name = tensor<string, []>("key_states_23_cast")]; | |
| tensor<int32, [3]> var_567 = const()[name = tensor<string, []>("op_567"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_32 = transpose(perm = var_552_perm_0, x = var_551_cast)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [12, 77, 64]> value_states_23_cast = reshape(shape = var_567, x = transpose_32)[name = tensor<string, []>("value_states_23_cast")]; | |
| tensor<int32, [3]> var_570_perm_0 = const()[name = tensor<string, []>("op_570_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_31_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_31_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_31 = transpose(perm = var_570_perm_0, x = key_states_23_cast)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_31_cast = matmul(transpose_x = attn_weights_31_transpose_x_0, transpose_y = attn_weights_31_transpose_y_0, x = query_states_11_cast, y = transpose_31)[name = tensor<string, []>("attn_weights_31_cast")]; | |
| tensor<int32, [4]> var_572 = const()[name = tensor<string, []>("op_572"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_573_cast = reshape(shape = var_572, x = attn_weights_31_cast)[name = tensor<string, []>("op_573_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_33_cast = add(x = var_573_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_33_cast")]; | |
| tensor<int32, [3]> var_578 = const()[name = tensor<string, []>("op_578"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_85_cast = reshape(shape = var_578, x = attn_weights_33_cast)[name = tensor<string, []>("input_85_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_87_cast = softmax(axis = var_5, x = input_85_cast)[name = tensor<string, []>("input_87_cast")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_31_cast = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast, y = value_states_23_cast)[name = tensor<string, []>("attn_output_31_cast")]; | |
| tensor<int32, [4]> var_583 = const()[name = tensor<string, []>("op_583"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast = reshape(shape = var_583, x = attn_output_31_cast)[name = tensor<string, []>("attn_output_33_cast")]; | |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_586 = const()[name = tensor<string, []>("op_586"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_30 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [1, 77, 768]> input_89_cast = reshape(shape = var_586, x = transpose_30)[name = tensor<string, []>("input_89_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103910592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353216)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_33_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast)[name = tensor<string, []>("hidden_states_33_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_91_cast = add(x = input_83_cast, y = hidden_states_33_cast)[name = tensor<string, []>("input_91_cast")]; | |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104354816)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104356416)))]; | |
| tensor<fp16, [1, 77, 768]> input_93_cast = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast)[name = tensor<string, []>("input_93_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104358016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_95_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast)[name = tensor<string, []>("input_95_cast")]; | |
| tensor<fp16, []> var_601_to_fp16 = const()[name = tensor<string, []>("op_601_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_602_cast = mul(x = input_95_cast, y = var_601_to_fp16)[name = tensor<string, []>("op_602_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_603_cast = sigmoid(x = var_602_cast)[name = tensor<string, []>("op_603_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_97_cast = mul(x = input_95_cast, y = var_603_cast)[name = tensor<string, []>("input_97_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107899840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107900032)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_35_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast)[name = tensor<string, []>("hidden_states_35_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_99_cast = add(x = input_91_cast, y = hidden_states_35_cast)[name = tensor<string, []>("input_99_cast")]; | |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107901632)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107903232)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_37_cast = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast)[name = tensor<string, []>("hidden_states_37_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107904832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108347264))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108347456)))]; | |
| tensor<fp16, [1, 77, 768]> var_627_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast)[name = tensor<string, []>("op_627_cast")]; | |
| tensor<fp16, []> var_628_to_fp16 = const()[name = tensor<string, []>("op_628_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_41_cast = mul(x = var_627_cast, y = var_628_to_fp16)[name = tensor<string, []>("tensor_41_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108349056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108791488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108791680)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_37_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast)[name = tensor<string, []>("tensor_37_cast")]; | |
| tensor<int32, [4]> var_633 = const()[name = tensor<string, []>("op_633"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_634_cast = reshape(shape = var_633, x = tensor_37_cast)[name = tensor<string, []>("op_634_cast")]; | |
| tensor<int32, [4]> var_635_perm_0 = const()[name = tensor<string, []>("op_635_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108793280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109235712))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109235904)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_39_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast)[name = tensor<string, []>("tensor_39_cast")]; | |
| tensor<int32, [4]> var_640 = const()[name = tensor<string, []>("op_640"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_641_cast = reshape(shape = var_640, x = tensor_39_cast)[name = tensor<string, []>("op_641_cast")]; | |
| tensor<int32, [4]> var_642_perm_0 = const()[name = tensor<string, []>("op_642_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_649 = const()[name = tensor<string, []>("op_649"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_650_cast = reshape(shape = var_649, x = tensor_41_cast)[name = tensor<string, []>("op_650_cast")]; | |
| tensor<int32, [4]> var_651_perm_0 = const()[name = tensor<string, []>("op_651_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_653 = const()[name = tensor<string, []>("op_653"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_29 = transpose(perm = var_651_perm_0, x = var_650_cast)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [12, 77, 64]> query_states_13_cast = reshape(shape = var_653, x = transpose_29)[name = tensor<string, []>("query_states_13_cast")]; | |
| tensor<int32, [3]> var_655 = const()[name = tensor<string, []>("op_655"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_28 = transpose(perm = var_635_perm_0, x = var_634_cast)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [12, 77, 64]> key_states_27_cast = reshape(shape = var_655, x = transpose_28)[name = tensor<string, []>("key_states_27_cast")]; | |
| tensor<int32, [3]> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_27 = transpose(perm = var_642_perm_0, x = var_641_cast)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [12, 77, 64]> value_states_27_cast = reshape(shape = var_657, x = transpose_27)[name = tensor<string, []>("value_states_27_cast")]; | |
| tensor<int32, [3]> var_660_perm_0 = const()[name = tensor<string, []>("op_660_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_37_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_37_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_26 = transpose(perm = var_660_perm_0, x = key_states_27_cast)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_37_cast = matmul(transpose_x = attn_weights_37_transpose_x_0, transpose_y = attn_weights_37_transpose_y_0, x = query_states_13_cast, y = transpose_26)[name = tensor<string, []>("attn_weights_37_cast")]; | |
| tensor<int32, [4]> var_662 = const()[name = tensor<string, []>("op_662"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_663_cast = reshape(shape = var_662, x = attn_weights_37_cast)[name = tensor<string, []>("op_663_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_39_cast = add(x = var_663_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_39_cast")]; | |
| tensor<int32, [3]> var_668 = const()[name = tensor<string, []>("op_668"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_101_cast = reshape(shape = var_668, x = attn_weights_39_cast)[name = tensor<string, []>("input_101_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_103_cast = softmax(axis = var_5, x = input_101_cast)[name = tensor<string, []>("input_103_cast")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_37_cast = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast, y = value_states_27_cast)[name = tensor<string, []>("attn_output_37_cast")]; | |
| tensor<int32, [4]> var_673 = const()[name = tensor<string, []>("op_673"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_39_cast = reshape(shape = var_673, x = attn_output_37_cast)[name = tensor<string, []>("attn_output_39_cast")]; | |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_25 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [1, 77, 768]> input_105_cast = reshape(shape = var_676, x = transpose_25)[name = tensor<string, []>("input_105_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109237504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109679936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109680128)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_39_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast)[name = tensor<string, []>("hidden_states_39_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_107_cast = add(x = input_99_cast, y = hidden_states_39_cast)[name = tensor<string, []>("input_107_cast")]; | |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109681728)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109683328)))]; | |
| tensor<fp16, [1, 77, 768]> input_109_cast = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast)[name = tensor<string, []>("input_109_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109684928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_111_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast)[name = tensor<string, []>("input_111_cast")]; | |
| tensor<fp16, []> var_691_to_fp16 = const()[name = tensor<string, []>("op_691_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_692_cast = mul(x = input_111_cast, y = var_691_to_fp16)[name = tensor<string, []>("op_692_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_693_cast = sigmoid(x = var_692_cast)[name = tensor<string, []>("op_693_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_113_cast = mul(x = input_111_cast, y = var_693_cast)[name = tensor<string, []>("input_113_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113226752))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113226944)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_41_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast)[name = tensor<string, []>("hidden_states_41_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_115_cast = add(x = input_107_cast, y = hidden_states_41_cast)[name = tensor<string, []>("input_115_cast")]; | |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113228544)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113230144)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_43_cast = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast)[name = tensor<string, []>("hidden_states_43_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113231744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113674176))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113674368)))]; | |
| tensor<fp16, [1, 77, 768]> var_717_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast)[name = tensor<string, []>("op_717_cast")]; | |
| tensor<fp16, []> var_718_to_fp16 = const()[name = tensor<string, []>("op_718_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_47_cast = mul(x = var_717_cast, y = var_718_to_fp16)[name = tensor<string, []>("tensor_47_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113675968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114118400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114118592)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_43_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast)[name = tensor<string, []>("tensor_43_cast")]; | |
| tensor<int32, [4]> var_723 = const()[name = tensor<string, []>("op_723"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_724_cast = reshape(shape = var_723, x = tensor_43_cast)[name = tensor<string, []>("op_724_cast")]; | |
| tensor<int32, [4]> var_725_perm_0 = const()[name = tensor<string, []>("op_725_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114120192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114562624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114562816)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_45_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast)[name = tensor<string, []>("tensor_45_cast")]; | |
| tensor<int32, [4]> var_730 = const()[name = tensor<string, []>("op_730"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_731_cast = reshape(shape = var_730, x = tensor_45_cast)[name = tensor<string, []>("op_731_cast")]; | |
| tensor<int32, [4]> var_732_perm_0 = const()[name = tensor<string, []>("op_732_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_739 = const()[name = tensor<string, []>("op_739"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_740_cast = reshape(shape = var_739, x = tensor_47_cast)[name = tensor<string, []>("op_740_cast")]; | |
| tensor<int32, [4]> var_741_perm_0 = const()[name = tensor<string, []>("op_741_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_743 = const()[name = tensor<string, []>("op_743"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_24 = transpose(perm = var_741_perm_0, x = var_740_cast)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [12, 77, 64]> query_states_15_cast = reshape(shape = var_743, x = transpose_24)[name = tensor<string, []>("query_states_15_cast")]; | |
| tensor<int32, [3]> var_745 = const()[name = tensor<string, []>("op_745"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_23 = transpose(perm = var_725_perm_0, x = var_724_cast)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [12, 77, 64]> key_states_31_cast = reshape(shape = var_745, x = transpose_23)[name = tensor<string, []>("key_states_31_cast")]; | |
| tensor<int32, [3]> var_747 = const()[name = tensor<string, []>("op_747"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_22 = transpose(perm = var_732_perm_0, x = var_731_cast)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [12, 77, 64]> value_states_31_cast = reshape(shape = var_747, x = transpose_22)[name = tensor<string, []>("value_states_31_cast")]; | |
| tensor<int32, [3]> var_750_perm_0 = const()[name = tensor<string, []>("op_750_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_43_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_43_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_21 = transpose(perm = var_750_perm_0, x = key_states_31_cast)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_43_cast = matmul(transpose_x = attn_weights_43_transpose_x_0, transpose_y = attn_weights_43_transpose_y_0, x = query_states_15_cast, y = transpose_21)[name = tensor<string, []>("attn_weights_43_cast")]; | |
| tensor<int32, [4]> var_752 = const()[name = tensor<string, []>("op_752"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_753_cast = reshape(shape = var_752, x = attn_weights_43_cast)[name = tensor<string, []>("op_753_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_45_cast = add(x = var_753_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_45_cast")]; | |
| tensor<int32, [3]> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_117_cast = reshape(shape = var_758, x = attn_weights_45_cast)[name = tensor<string, []>("input_117_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_119_cast = softmax(axis = var_5, x = input_117_cast)[name = tensor<string, []>("input_119_cast")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_43_cast = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast, y = value_states_31_cast)[name = tensor<string, []>("attn_output_43_cast")]; | |
| tensor<int32, [4]> var_763 = const()[name = tensor<string, []>("op_763"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast = reshape(shape = var_763, x = attn_output_43_cast)[name = tensor<string, []>("attn_output_45_cast")]; | |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_766 = const()[name = tensor<string, []>("op_766"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_20 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [1, 77, 768]> input_121_cast = reshape(shape = var_766, x = transpose_20)[name = tensor<string, []>("input_121_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114564416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115006848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115007040)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_45_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast)[name = tensor<string, []>("hidden_states_45_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_123_cast = add(x = input_115_cast, y = hidden_states_45_cast)[name = tensor<string, []>("input_123_cast")]; | |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115008640)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115010240)))]; | |
| tensor<fp16, [1, 77, 768]> input_125_cast = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast)[name = tensor<string, []>("input_125_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115011840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116783936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_127_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast)[name = tensor<string, []>("input_127_cast")]; | |
| tensor<fp16, []> var_781_to_fp16 = const()[name = tensor<string, []>("op_781_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_782_cast = mul(x = input_127_cast, y = var_781_to_fp16)[name = tensor<string, []>("op_782_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_783_cast = sigmoid(x = var_782_cast)[name = tensor<string, []>("op_783_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_129_cast = mul(x = input_127_cast, y = var_783_cast)[name = tensor<string, []>("input_129_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116784128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553664))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553856)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_47_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast)[name = tensor<string, []>("hidden_states_47_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_131_cast = add(x = input_123_cast, y = hidden_states_47_cast)[name = tensor<string, []>("input_131_cast")]; | |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118555456)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118557056)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_49_cast = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast)[name = tensor<string, []>("hidden_states_49_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118558656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119001088))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119001280)))]; | |
| tensor<fp16, [1, 77, 768]> var_807_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast)[name = tensor<string, []>("op_807_cast")]; | |
| tensor<fp16, []> var_808_to_fp16 = const()[name = tensor<string, []>("op_808_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_53_cast = mul(x = var_807_cast, y = var_808_to_fp16)[name = tensor<string, []>("tensor_53_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119002880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119445312))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119445504)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_49_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast)[name = tensor<string, []>("tensor_49_cast")]; | |
| tensor<int32, [4]> var_813 = const()[name = tensor<string, []>("op_813"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_814_cast = reshape(shape = var_813, x = tensor_49_cast)[name = tensor<string, []>("op_814_cast")]; | |
| tensor<int32, [4]> var_815_perm_0 = const()[name = tensor<string, []>("op_815_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119447104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119889536))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119889728)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_51_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast)[name = tensor<string, []>("tensor_51_cast")]; | |
| tensor<int32, [4]> var_820 = const()[name = tensor<string, []>("op_820"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_821_cast = reshape(shape = var_820, x = tensor_51_cast)[name = tensor<string, []>("op_821_cast")]; | |
| tensor<int32, [4]> var_822_perm_0 = const()[name = tensor<string, []>("op_822_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_829 = const()[name = tensor<string, []>("op_829"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_830_cast = reshape(shape = var_829, x = tensor_53_cast)[name = tensor<string, []>("op_830_cast")]; | |
| tensor<int32, [4]> var_831_perm_0 = const()[name = tensor<string, []>("op_831_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_833 = const()[name = tensor<string, []>("op_833"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_19 = transpose(perm = var_831_perm_0, x = var_830_cast)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [12, 77, 64]> query_states_17_cast = reshape(shape = var_833, x = transpose_19)[name = tensor<string, []>("query_states_17_cast")]; | |
| tensor<int32, [3]> var_835 = const()[name = tensor<string, []>("op_835"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_18 = transpose(perm = var_815_perm_0, x = var_814_cast)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [12, 77, 64]> key_states_35_cast = reshape(shape = var_835, x = transpose_18)[name = tensor<string, []>("key_states_35_cast")]; | |
| tensor<int32, [3]> var_837 = const()[name = tensor<string, []>("op_837"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_17 = transpose(perm = var_822_perm_0, x = var_821_cast)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [12, 77, 64]> value_states_35_cast = reshape(shape = var_837, x = transpose_17)[name = tensor<string, []>("value_states_35_cast")]; | |
| tensor<int32, [3]> var_840_perm_0 = const()[name = tensor<string, []>("op_840_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_49_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_49_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_16 = transpose(perm = var_840_perm_0, x = key_states_35_cast)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_49_cast = matmul(transpose_x = attn_weights_49_transpose_x_0, transpose_y = attn_weights_49_transpose_y_0, x = query_states_17_cast, y = transpose_16)[name = tensor<string, []>("attn_weights_49_cast")]; | |
| tensor<int32, [4]> var_842 = const()[name = tensor<string, []>("op_842"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_843_cast = reshape(shape = var_842, x = attn_weights_49_cast)[name = tensor<string, []>("op_843_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_51_cast = add(x = var_843_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_51_cast")]; | |
| tensor<int32, [3]> var_848 = const()[name = tensor<string, []>("op_848"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_133_cast = reshape(shape = var_848, x = attn_weights_51_cast)[name = tensor<string, []>("input_133_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_135_cast = softmax(axis = var_5, x = input_133_cast)[name = tensor<string, []>("input_135_cast")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_49_cast = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast, y = value_states_35_cast)[name = tensor<string, []>("attn_output_49_cast")]; | |
| tensor<int32, [4]> var_853 = const()[name = tensor<string, []>("op_853"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_51_cast = reshape(shape = var_853, x = attn_output_49_cast)[name = tensor<string, []>("attn_output_51_cast")]; | |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_856 = const()[name = tensor<string, []>("op_856"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_15 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [1, 77, 768]> input_137_cast = reshape(shape = var_856, x = transpose_15)[name = tensor<string, []>("input_137_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119891328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120333760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120333952)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_51_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast)[name = tensor<string, []>("hidden_states_51_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_139_cast = add(x = input_131_cast, y = hidden_states_51_cast)[name = tensor<string, []>("input_139_cast")]; | |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120335552)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120337152)))]; | |
| tensor<fp16, [1, 77, 768]> input_141_cast = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast)[name = tensor<string, []>("input_141_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120338752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122110848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_143_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast)[name = tensor<string, []>("input_143_cast")]; | |
| tensor<fp16, []> var_871_to_fp16 = const()[name = tensor<string, []>("op_871_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_872_cast = mul(x = input_143_cast, y = var_871_to_fp16)[name = tensor<string, []>("op_872_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_873_cast = sigmoid(x = var_872_cast)[name = tensor<string, []>("op_873_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_145_cast = mul(x = input_143_cast, y = var_873_cast)[name = tensor<string, []>("input_145_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122111040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123880576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123880768)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_53_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast)[name = tensor<string, []>("hidden_states_53_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_147_cast = add(x = input_139_cast, y = hidden_states_53_cast)[name = tensor<string, []>("input_147_cast")]; | |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123882368)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123883968)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_55_cast = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast)[name = tensor<string, []>("hidden_states_55_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123885568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124328000))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124328192)))]; | |
| tensor<fp16, [1, 77, 768]> var_897_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast)[name = tensor<string, []>("op_897_cast")]; | |
| tensor<fp16, []> var_898_to_fp16 = const()[name = tensor<string, []>("op_898_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_59_cast = mul(x = var_897_cast, y = var_898_to_fp16)[name = tensor<string, []>("tensor_59_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124329792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124772224))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124772416)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_55_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast)[name = tensor<string, []>("tensor_55_cast")]; | |
| tensor<int32, [4]> var_903 = const()[name = tensor<string, []>("op_903"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_904_cast = reshape(shape = var_903, x = tensor_55_cast)[name = tensor<string, []>("op_904_cast")]; | |
| tensor<int32, [4]> var_905_perm_0 = const()[name = tensor<string, []>("op_905_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124774016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125216448))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125216640)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_57_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast)[name = tensor<string, []>("tensor_57_cast")]; | |
| tensor<int32, [4]> var_910 = const()[name = tensor<string, []>("op_910"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_911_cast = reshape(shape = var_910, x = tensor_57_cast)[name = tensor<string, []>("op_911_cast")]; | |
| tensor<int32, [4]> var_912_perm_0 = const()[name = tensor<string, []>("op_912_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_919 = const()[name = tensor<string, []>("op_919"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_920_cast = reshape(shape = var_919, x = tensor_59_cast)[name = tensor<string, []>("op_920_cast")]; | |
| tensor<int32, [4]> var_921_perm_0 = const()[name = tensor<string, []>("op_921_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_923 = const()[name = tensor<string, []>("op_923"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_14 = transpose(perm = var_921_perm_0, x = var_920_cast)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [12, 77, 64]> query_states_19_cast = reshape(shape = var_923, x = transpose_14)[name = tensor<string, []>("query_states_19_cast")]; | |
| tensor<int32, [3]> var_925 = const()[name = tensor<string, []>("op_925"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_13 = transpose(perm = var_905_perm_0, x = var_904_cast)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [12, 77, 64]> key_states_39_cast = reshape(shape = var_925, x = transpose_13)[name = tensor<string, []>("key_states_39_cast")]; | |
| tensor<int32, [3]> var_927 = const()[name = tensor<string, []>("op_927"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_12 = transpose(perm = var_912_perm_0, x = var_911_cast)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [12, 77, 64]> value_states_39_cast = reshape(shape = var_927, x = transpose_12)[name = tensor<string, []>("value_states_39_cast")]; | |
| tensor<int32, [3]> var_930_perm_0 = const()[name = tensor<string, []>("op_930_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_55_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_55_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_11 = transpose(perm = var_930_perm_0, x = key_states_39_cast)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_55_cast = matmul(transpose_x = attn_weights_55_transpose_x_0, transpose_y = attn_weights_55_transpose_y_0, x = query_states_19_cast, y = transpose_11)[name = tensor<string, []>("attn_weights_55_cast")]; | |
| tensor<int32, [4]> var_932 = const()[name = tensor<string, []>("op_932"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_933_cast = reshape(shape = var_932, x = attn_weights_55_cast)[name = tensor<string, []>("op_933_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_57_cast = add(x = var_933_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_57_cast")]; | |
| tensor<int32, [3]> var_938 = const()[name = tensor<string, []>("op_938"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_149_cast = reshape(shape = var_938, x = attn_weights_57_cast)[name = tensor<string, []>("input_149_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_151_cast = softmax(axis = var_5, x = input_149_cast)[name = tensor<string, []>("input_151_cast")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_55_cast = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast, y = value_states_39_cast)[name = tensor<string, []>("attn_output_55_cast")]; | |
| tensor<int32, [4]> var_943 = const()[name = tensor<string, []>("op_943"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_57_cast = reshape(shape = var_943, x = attn_output_55_cast)[name = tensor<string, []>("attn_output_57_cast")]; | |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_946 = const()[name = tensor<string, []>("op_946"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_10 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [1, 77, 768]> input_153_cast = reshape(shape = var_946, x = transpose_10)[name = tensor<string, []>("input_153_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125218240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125660672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125660864)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_57_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast)[name = tensor<string, []>("hidden_states_57_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_155_cast = add(x = input_147_cast, y = hidden_states_57_cast)[name = tensor<string, []>("input_155_cast")]; | |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125662464)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125664064)))]; | |
| tensor<fp16, [1, 77, 768]> input_157_cast = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast)[name = tensor<string, []>("input_157_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125665664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_159_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast)[name = tensor<string, []>("input_159_cast")]; | |
| tensor<fp16, []> var_961_to_fp16 = const()[name = tensor<string, []>("op_961_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_962_cast = mul(x = input_159_cast, y = var_961_to_fp16)[name = tensor<string, []>("op_962_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_963_cast = sigmoid(x = var_962_cast)[name = tensor<string, []>("op_963_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_161_cast = mul(x = input_159_cast, y = var_963_cast)[name = tensor<string, []>("input_161_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129207488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129207680)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_59_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast)[name = tensor<string, []>("hidden_states_59_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_163_cast = add(x = input_155_cast, y = hidden_states_59_cast)[name = tensor<string, []>("input_163_cast")]; | |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129209280)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129210880)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_61_cast = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast)[name = tensor<string, []>("hidden_states_61_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129212480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129654912))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129655104)))]; | |
| tensor<fp16, [1, 77, 768]> var_987_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast)[name = tensor<string, []>("op_987_cast")]; | |
| tensor<fp16, []> var_988_to_fp16 = const()[name = tensor<string, []>("op_988_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_65_cast = mul(x = var_987_cast, y = var_988_to_fp16)[name = tensor<string, []>("tensor_65_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129656704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130099136))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130099328)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_61_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast)[name = tensor<string, []>("tensor_61_cast")]; | |
| tensor<int32, [4]> var_993 = const()[name = tensor<string, []>("op_993"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_994_cast = reshape(shape = var_993, x = tensor_61_cast)[name = tensor<string, []>("op_994_cast")]; | |
| tensor<int32, [4]> var_995_perm_0 = const()[name = tensor<string, []>("op_995_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130100928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130543360))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130543552)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_63_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast)[name = tensor<string, []>("tensor_63_cast")]; | |
| tensor<int32, [4]> var_1000 = const()[name = tensor<string, []>("op_1000"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1001_cast = reshape(shape = var_1000, x = tensor_63_cast)[name = tensor<string, []>("op_1001_cast")]; | |
| tensor<int32, [4]> var_1002_perm_0 = const()[name = tensor<string, []>("op_1002_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1009 = const()[name = tensor<string, []>("op_1009"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1010_cast = reshape(shape = var_1009, x = tensor_65_cast)[name = tensor<string, []>("op_1010_cast")]; | |
| tensor<int32, [4]> var_1011_perm_0 = const()[name = tensor<string, []>("op_1011_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1013 = const()[name = tensor<string, []>("op_1013"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_9 = transpose(perm = var_1011_perm_0, x = var_1010_cast)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [12, 77, 64]> query_states_21_cast = reshape(shape = var_1013, x = transpose_9)[name = tensor<string, []>("query_states_21_cast")]; | |
| tensor<int32, [3]> var_1015 = const()[name = tensor<string, []>("op_1015"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_8 = transpose(perm = var_995_perm_0, x = var_994_cast)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [12, 77, 64]> key_states_43_cast = reshape(shape = var_1015, x = transpose_8)[name = tensor<string, []>("key_states_43_cast")]; | |
| tensor<int32, [3]> var_1017 = const()[name = tensor<string, []>("op_1017"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_7 = transpose(perm = var_1002_perm_0, x = var_1001_cast)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [12, 77, 64]> value_states_43_cast = reshape(shape = var_1017, x = transpose_7)[name = tensor<string, []>("value_states_43_cast")]; | |
| tensor<int32, [3]> var_1020_perm_0 = const()[name = tensor<string, []>("op_1020_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_61_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_61_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_6 = transpose(perm = var_1020_perm_0, x = key_states_43_cast)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_61_cast = matmul(transpose_x = attn_weights_61_transpose_x_0, transpose_y = attn_weights_61_transpose_y_0, x = query_states_21_cast, y = transpose_6)[name = tensor<string, []>("attn_weights_61_cast")]; | |
| tensor<int32, [4]> var_1022 = const()[name = tensor<string, []>("op_1022"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1023_cast = reshape(shape = var_1022, x = attn_weights_61_cast)[name = tensor<string, []>("op_1023_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_63_cast = add(x = var_1023_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_63_cast")]; | |
| tensor<int32, [3]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_165_cast = reshape(shape = var_1028, x = attn_weights_63_cast)[name = tensor<string, []>("input_165_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_167_cast = softmax(axis = var_5, x = input_165_cast)[name = tensor<string, []>("input_167_cast")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_61_cast = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast, y = value_states_43_cast)[name = tensor<string, []>("attn_output_61_cast")]; | |
| tensor<int32, [4]> var_1033 = const()[name = tensor<string, []>("op_1033"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_63_cast = reshape(shape = var_1033, x = attn_output_61_cast)[name = tensor<string, []>("attn_output_63_cast")]; | |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1036 = const()[name = tensor<string, []>("op_1036"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [1, 77, 768]> input_169_cast = reshape(shape = var_1036, x = transpose_5)[name = tensor<string, []>("input_169_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130545152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130987584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130987776)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_63_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast)[name = tensor<string, []>("hidden_states_63_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_171_cast = add(x = input_163_cast, y = hidden_states_63_cast)[name = tensor<string, []>("input_171_cast")]; | |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130989376)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130990976)))]; | |
| tensor<fp16, [1, 77, 768]> input_173_cast = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast)[name = tensor<string, []>("input_173_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130992576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_175_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast)[name = tensor<string, []>("input_175_cast")]; | |
| tensor<fp16, []> var_1051_to_fp16 = const()[name = tensor<string, []>("op_1051_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1052_cast = mul(x = input_175_cast, y = var_1051_to_fp16)[name = tensor<string, []>("op_1052_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_1053_cast = sigmoid(x = var_1052_cast)[name = tensor<string, []>("op_1053_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_177_cast = mul(x = input_175_cast, y = var_1053_cast)[name = tensor<string, []>("input_177_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134534400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134534592)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_65_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast)[name = tensor<string, []>("hidden_states_65_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_179_cast = add(x = input_171_cast, y = hidden_states_65_cast)[name = tensor<string, []>("input_179_cast")]; | |
| tensor<string, []> input_179_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("input_179_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134536192)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134537792)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_67_cast = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast)[name = tensor<string, []>("hidden_states_67_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134539392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134981824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134982016)))]; | |
| tensor<fp16, [1, 77, 768]> var_1077_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast)[name = tensor<string, []>("op_1077_cast")]; | |
| tensor<fp16, []> var_1078_to_fp16 = const()[name = tensor<string, []>("op_1078_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_cast = mul(x = var_1077_cast, y = var_1078_to_fp16)[name = tensor<string, []>("tensor_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134983616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135426048))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135426240)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_67_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast)[name = tensor<string, []>("tensor_67_cast")]; | |
| tensor<int32, [4]> var_1083 = const()[name = tensor<string, []>("op_1083"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1084_cast = reshape(shape = var_1083, x = tensor_67_cast)[name = tensor<string, []>("op_1084_cast")]; | |
| tensor<int32, [4]> var_1085_perm_0 = const()[name = tensor<string, []>("op_1085_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135427840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135870272))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135870464)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_69_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast)[name = tensor<string, []>("tensor_69_cast")]; | |
| tensor<int32, [4]> var_1090 = const()[name = tensor<string, []>("op_1090"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1091_cast = reshape(shape = var_1090, x = tensor_69_cast)[name = tensor<string, []>("op_1091_cast")]; | |
| tensor<int32, [4]> var_1092_perm_0 = const()[name = tensor<string, []>("op_1092_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1099 = const()[name = tensor<string, []>("op_1099"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1100_cast = reshape(shape = var_1099, x = tensor_cast)[name = tensor<string, []>("op_1100_cast")]; | |
| tensor<int32, [4]> var_1101_perm_0 = const()[name = tensor<string, []>("op_1101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1103 = const()[name = tensor<string, []>("op_1103"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_4 = transpose(perm = var_1101_perm_0, x = var_1100_cast)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [12, 77, 64]> query_states_cast = reshape(shape = var_1103, x = transpose_4)[name = tensor<string, []>("query_states_cast")]; | |
| tensor<int32, [3]> var_1105 = const()[name = tensor<string, []>("op_1105"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_3 = transpose(perm = var_1085_perm_0, x = var_1084_cast)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [12, 77, 64]> key_states_cast = reshape(shape = var_1105, x = transpose_3)[name = tensor<string, []>("key_states_cast")]; | |
| tensor<int32, [3]> var_1107 = const()[name = tensor<string, []>("op_1107"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_2 = transpose(perm = var_1092_perm_0, x = var_1091_cast)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [12, 77, 64]> value_states_cast = reshape(shape = var_1107, x = transpose_2)[name = tensor<string, []>("value_states_cast")]; | |
| tensor<int32, [3]> var_1110_perm_0 = const()[name = tensor<string, []>("op_1110_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_67_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_67_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_1 = transpose(perm = var_1110_perm_0, x = key_states_cast)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_67_cast = matmul(transpose_x = attn_weights_67_transpose_x_0, transpose_y = attn_weights_67_transpose_y_0, x = query_states_cast, y = transpose_1)[name = tensor<string, []>("attn_weights_67_cast")]; | |
| tensor<int32, [4]> var_1112 = const()[name = tensor<string, []>("op_1112"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1113_cast = reshape(shape = var_1112, x = attn_weights_67_cast)[name = tensor<string, []>("op_1113_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_69_cast = add(x = var_1113_cast, y = causal_attention_mask_to_fp16_palettized)[name = tensor<string, []>("attn_weights_69_cast")]; | |
| tensor<int32, [3]> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_181_cast = reshape(shape = var_1118, x = attn_weights_69_cast)[name = tensor<string, []>("input_181_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_183_cast = softmax(axis = var_5, x = input_181_cast)[name = tensor<string, []>("input_183_cast")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_67_cast = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast, y = value_states_cast)[name = tensor<string, []>("attn_output_67_cast")]; | |
| tensor<int32, [4]> var_1123 = const()[name = tensor<string, []>("op_1123"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_69_cast = reshape(shape = var_1123, x = attn_output_67_cast)[name = tensor<string, []>("attn_output_69_cast")]; | |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1126 = const()[name = tensor<string, []>("op_1126"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast)[name = tensor<string, []>("transpose_0")]; | |
| tensor<fp16, [1, 77, 768]> input_185_cast = reshape(shape = var_1126, x = transpose_0)[name = tensor<string, []>("input_185_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135872064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136314496))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136314688)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_69_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast)[name = tensor<string, []>("hidden_states_69_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_187_cast = add(x = input_179_cast, y = hidden_states_69_cast)[name = tensor<string, []>("input_187_cast")]; | |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136316288)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136317888)))]; | |
| tensor<fp16, [1, 77, 768]> input_189_cast = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast)[name = tensor<string, []>("input_189_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136319488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_191_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast)[name = tensor<string, []>("input_191_cast")]; | |
| tensor<fp16, []> var_1141_to_fp16 = const()[name = tensor<string, []>("op_1141_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1142_cast = mul(x = input_191_cast, y = var_1141_to_fp16)[name = tensor<string, []>("op_1142_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_1143_cast = sigmoid(x = var_1142_cast)[name = tensor<string, []>("op_1143_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_193_cast = mul(x = input_191_cast, y = var_1143_cast)[name = tensor<string, []>("input_193_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139861312))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139861504)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast)[name = tensor<string, []>("hidden_states_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_cast = add(x = input_187_cast, y = hidden_states_cast)[name = tensor<string, []>("input_cast")]; | |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139863104)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139864704)))]; | |
| tensor<fp16, [1, 77, 768]> last_hidden_state_cast = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_13_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast)[name = tensor<string, []>("last_hidden_state_cast")]; | |
| tensor<string, []> last_hidden_state_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("last_hidden_state_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 77, 768]> last_hidden_state = cast(dtype = input_179_cast_to_fp32_dtype_0, x = input_179_cast)[name = tensor<string, []>("cast_0")]; | |
| tensor<fp32, [1, 77, 768]> pooled_outputs = cast(dtype = last_hidden_state_cast_to_fp32_dtype_0, x = last_hidden_state_cast)[name = tensor<string, []>("cast_1")]; | |
| } -> (last_hidden_state, pooled_outputs); | |
| } |