program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.10.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios17>(tensor<int32, [1, 512]> attention_mask, tensor<int32, [1, 512]> input_ids) {
            tensor<fp32, [50265, 768]> text_branch_embeddings_word_embeddings_weight = const()[name = tensor<string, []>("text_branch_embeddings_word_embeddings_weight"), val = tensor<fp32, [50265, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp32, [514, 768]> text_branch_embeddings_position_embeddings_weight = const()[name = tensor<string, []>("text_branch_embeddings_position_embeddings_weight"), val = tensor<fp32, [514, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154414208)))];
            tensor<fp32, [768]> text_branch_embeddings_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_embeddings_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155993280)))];
            tensor<fp32, [768]> text_branch_embeddings_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_embeddings_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155996416)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155999552)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_0_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156002688)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158362048)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_0_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158365184)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160724544)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_0_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160727680)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163087040)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_0_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163090176)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165449536)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165452672)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_0_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165455808)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_0_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165468160)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174905408)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_0_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174908544)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_0_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184345792)))];
            tensor<fp32, [768]> text_branch_encoder_layer_0_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_0_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184348928)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184352064)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_1_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184355200)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186714560)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_1_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186717696)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189077056)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_1_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189080192)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191439552)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_1_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191442688)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193802048)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193805184)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_1_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193808320)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_1_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193820672)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203257920)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_1_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203261056)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_1_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212698304)))];
            tensor<fp32, [768]> text_branch_encoder_layer_1_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_1_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212701440)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212704576)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_2_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212707712)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215067072)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_2_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215070208)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217429568)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_2_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217432704)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219792064)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_2_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219795200)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222154560)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222157696)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_2_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222160832)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_2_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222173184)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231610432)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_2_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231613568)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_2_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241050816)))];
            tensor<fp32, [768]> text_branch_encoder_layer_2_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_2_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241053952)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241057088)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_3_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241060224)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243419584)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_3_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243422720)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245782080)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_3_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245785216)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248144576)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_3_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248147712)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250507072)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250510208)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_3_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250513344)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_3_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250525696)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259962944)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_3_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259966080)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_3_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269403328)))];
            tensor<fp32, [768]> text_branch_encoder_layer_3_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_3_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269406464)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269409600)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_4_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269412736)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271772096)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_4_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271775232)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274134592)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_4_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274137728)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276497088)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_4_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276500224)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278859584)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278862720)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_4_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278865856)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_4_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278878208)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288315456)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_4_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288318592)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_4_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297755840)))];
            tensor<fp32, [768]> text_branch_encoder_layer_4_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_4_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297758976)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297762112)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_5_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297765248)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300124608)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_5_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300127744)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302487104)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_5_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302490240)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304849600)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_5_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304852736)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307212096)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307215232)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_5_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307218368)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_5_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307230720)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316667968)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_5_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316671104)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_5_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326108352)))];
            tensor<fp32, [768]> text_branch_encoder_layer_5_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_5_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326111488)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326114624)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_6_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326117760)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328477120)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_6_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328480256)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330839616)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_6_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330842752)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333202112)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_6_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333205248)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335564608)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335567744)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_6_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335570880)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_6_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335583232)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345020480)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_6_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345023616)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_6_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354460864)))];
            tensor<fp32, [768]> text_branch_encoder_layer_6_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_6_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354464000)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354467136)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_7_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354470272)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356829632)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_7_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356832768)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(359192128)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_7_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(359195264)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361554624)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_7_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361557760)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363917120)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363920256)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_7_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363923392)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_7_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363935744)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373372992)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_7_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373376128)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_7_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382813376)))];
            tensor<fp32, [768]> text_branch_encoder_layer_7_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_7_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382816512)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382819648)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_8_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382822784)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385182144)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_8_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385185280)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387544640)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_8_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387547776)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389907136)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_8_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389910272)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392269632)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392272768)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_8_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392275904)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_8_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(392288256)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401725504)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_8_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401728640)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_8_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411165888)))];
            tensor<fp32, [768]> text_branch_encoder_layer_8_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_8_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411169024)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411172160)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_9_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411175296)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413534656)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_9_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413537792)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415897152)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_9_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415900288)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418259648)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_9_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418262784)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420622144)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420625280)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_9_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420628416)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_9_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420640768)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430078016)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_9_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430081152)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_9_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439518400)))];
            tensor<fp32, [768]> text_branch_encoder_layer_9_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_9_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439521536)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439524672)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_10_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439527808)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441887168)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_10_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441890304)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444249664)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_10_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444252800)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446612160)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_10_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446615296)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448974656)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448977792)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_10_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448980928)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_10_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448993280)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(458430528)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_10_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(458433664)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_10_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467870912)))];
            tensor<fp32, [768]> text_branch_encoder_layer_10_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_10_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467874048)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_self_query_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_query_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467877184)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_11_attention_self_query_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_query_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467880320)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_self_key_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_key_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470239680)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_11_attention_self_key_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_key_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470242816)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_self_value_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_value_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(472602176)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_11_attention_self_value_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_self_value_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(472605312)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(474964672)))];
            tensor<fp32, [768, 768]> text_branch_encoder_layer_11_attention_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_output_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(474967808)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477327168)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_attention_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_attention_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477330304)))];
            tensor<fp32, [3072]> text_branch_encoder_layer_11_intermediate_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_intermediate_dense_bias"), val = tensor<fp32, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477333440)))];
            tensor<fp32, [3072, 768]> text_branch_encoder_layer_11_intermediate_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_intermediate_dense_weight"), val = tensor<fp32, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477345792)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_output_dense_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_output_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486783040)))];
            tensor<fp32, [768, 3072]> text_branch_encoder_layer_11_output_dense_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_output_dense_weight"), val = tensor<fp32, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486786176)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_output_LayerNorm_bias = const()[name = tensor<string, []>("text_branch_encoder_layer_11_output_LayerNorm_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496223424)))];
            tensor<fp32, [768]> text_branch_encoder_layer_11_output_LayerNorm_weight = const()[name = tensor<string, []>("text_branch_encoder_layer_11_output_LayerNorm_weight"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496226560)))];
            tensor<fp32, [768]> text_branch_pooler_dense_bias = const()[name = tensor<string, []>("text_branch_pooler_dense_bias"), val = tensor<fp32, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496229696)))];
            tensor<fp32, [768, 768]> text_branch_pooler_dense_weight = const()[name = tensor<string, []>("text_branch_pooler_dense_weight"), val = tensor<fp32, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496232832)))];
            tensor<fp32, [512]> text_projection_0_bias = const()[name = tensor<string, []>("text_projection_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498592192)))];
            tensor<fp32, [512, 768]> text_projection_0_weight = const()[name = tensor<string, []>("text_projection_0_weight"), val = tensor<fp32, [512, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498594304)))];
            tensor<fp32, [512]> text_projection_2_bias = const()[name = tensor<string, []>("text_projection_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500167232)))];
            tensor<fp32, [512, 512]> text_projection_2_weight = const()[name = tensor<string, []>("text_projection_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500169344)))];
            tensor<int32, []> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<int32, []>(-1)];
            tensor<fp32, []> var_12 = const()[name = tensor<string, []>("op_12"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
            tensor<fp32, []> var_15 = const()[name = tensor<string, []>("op_15"), val = tensor<fp32, []>(0x1p+0)];
            tensor<int32, []> var_22 = const()[name = tensor<string, []>("op_22"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_36_axes_0 = const()[name = tensor<string, []>("op_36_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1, 512]> var_36 = expand_dims(axes = var_36_axes_0, x = attention_mask)[name = tensor<string, []>("op_36")];
            tensor<int32, [1]> var_37_axes_0 = const()[name = tensor<string, []>("op_37_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1, 1, 1, 512]> var_37 = expand_dims(axes = var_37_axes_0, x = var_36)[name = tensor<string, []>("op_37")];
            tensor<string, []> var_39_dtype_0 = const()[name = tensor<string, []>("op_39_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 1, 1, 512]> var_39 = cast(dtype = var_39_dtype_0, x = var_37)[name = tensor<string, []>("cast_78")];
            tensor<fp32, [1, 1, 1, 512]> var_40 = sub(x = var_15, y = var_39)[name = tensor<string, []>("op_40")];
            tensor<fp32, []> var_41 = const()[name = tensor<string, []>("op_41"), val = tensor<fp32, []>(-0x1.fffffep+127)];
            tensor<fp32, [1, 1, 1, 512]> attention_mask_1 = mul(x = var_40, y = var_41)[name = tensor<string, []>("attention_mask")];
            tensor<bool, [1, 512]> var_47 = not_equal(x = input_ids, y = var_22)[name = tensor<string, []>("op_47")];
            tensor<string, []> mask_dtype_0 = const()[name = tensor<string, []>("mask_dtype_0"), val = tensor<string, []>("int32")];
            tensor<bool, []> var_49_exclusive_0 = const()[name = tensor<string, []>("op_49_exclusive_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_49_reverse_0 = const()[name = tensor<string, []>("op_49_reverse_0"), val = tensor<bool, []>(false)];
            tensor<int32, [1, 512]> mask = cast(dtype = mask_dtype_0, x = var_47)[name = tensor<string, []>("cast_77")];
            tensor<int32, [1, 512]> var_49 = cumsum(axis = var_22, exclusive = var_49_exclusive_0, reverse = var_49_reverse_0, x = mask)[name = tensor<string, []>("op_49")];
            tensor<int32, [1, 512]> incremental_indices = mul(x = var_49, y = mask)[name = tensor<string, []>("incremental_indices")];
            tensor<int32, []> var_55 = const()[name = tensor<string, []>("op_55"), val = tensor<int32, []>(1)];
            tensor<int32, [1, 512]> input_3 = add(x = incremental_indices, y = var_55)[name = tensor<string, []>("input_3")];
            tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> inputs_embeds_validate_indices_0 = const()[name = tensor<string, []>("inputs_embeds_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
            tensor<bool, [1, 512]> greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = tensor<string, []>("greater_equal_0")];
            tensor<int32, []> slice_by_index_0 = const()[name = tensor<string, []>("slice_by_index_0"), val = tensor<int32, []>(50265)];
            tensor<int32, [1, 512]> add_0 = add(x = input_ids, y = slice_by_index_0)[name = tensor<string, []>("add_0")];
            tensor<int32, [1, 512]> select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
            tensor<int32, []> inputs_embeds_axis_1 = const()[name = tensor<string, []>("inputs_embeds_axis_1"), val = tensor<int32, []>(0)];
            tensor<fp32, [1, 512, 768]> inputs_embeds = gather(axis = inputs_embeds_axis_1, batch_dims = inputs_embeds_batch_dims_0, indices = select_0, validate_indices = inputs_embeds_validate_indices_0, x = text_branch_embeddings_word_embeddings_weight)[name = tensor<string, []>("inputs_embeds")];
            tensor<fp32, [1, 512, 768]> token_type_embeddings_1 = const()[name = tensor<string, []>("token_type_embeddings_1"), val = tensor<fp32, [1, 512, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(501217984)))];
            tensor<fp32, [1, 512, 768]> embeddings_1 = add(x = inputs_embeds, y = token_type_embeddings_1)[name = tensor<string, []>("embeddings_1")];
            tensor<int32, []> position_embeddings_1_batch_dims_0 = const()[name = tensor<string, []>("position_embeddings_1_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> position_embeddings_1_validate_indices_0 = const()[name = tensor<string, []>("position_embeddings_1_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<int32, []> greater_equal_1_y_0 = const()[name = tensor<string, []>("greater_equal_1_y_0"), val = tensor<int32, []>(0)];
            tensor<bool, [1, 512]> greater_equal_1 = greater_equal(x = input_3, y = greater_equal_1_y_0)[name = tensor<string, []>("greater_equal_1")];
            tensor<int32, []> slice_by_index_1 = const()[name = tensor<string, []>("slice_by_index_1"), val = tensor<int32, []>(514)];
            tensor<int32, [1, 512]> add_1 = add(x = input_3, y = slice_by_index_1)[name = tensor<string, []>("add_1")];
            tensor<int32, [1, 512]> select_1 = select(a = input_3, b = add_1, cond = greater_equal_1)[name = tensor<string, []>("select_1")];
            tensor<int32, []> position_embeddings_1_axis_1 = const()[name = tensor<string, []>("position_embeddings_1_axis_1"), val = tensor<int32, []>(0)];
            tensor<fp32, [1, 512, 768]> position_embeddings_1 = gather(axis = position_embeddings_1_axis_1, batch_dims = position_embeddings_1_batch_dims_0, indices = select_1, validate_indices = position_embeddings_1_validate_indices_0, x = text_branch_embeddings_position_embeddings_weight)[name = tensor<string, []>("position_embeddings_1")];
            tensor<fp32, [1, 512, 768]> input_5 = add(x = embeddings_1, y = position_embeddings_1)[name = tensor<string, []>("input_5")];
            tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_7 = layer_norm(axes = input_7_axes_0, beta = text_branch_embeddings_LayerNorm_bias, epsilon = var_12, gamma = text_branch_embeddings_LayerNorm_weight, x = input_5)[name = tensor<string, []>("input_7")];
            tensor<fp32, [1, 512, 768]> x_9 = linear(bias = text_branch_encoder_layer_0_attention_self_query_bias, weight = text_branch_encoder_layer_0_attention_self_query_weight, x = input_7)[name = tensor<string, []>("linear_0")];
            tensor<fp32, [1, 512, 768]> x_1 = linear(bias = text_branch_encoder_layer_0_attention_self_key_bias, weight = text_branch_encoder_layer_0_attention_self_key_weight, x = input_7)[name = tensor<string, []>("linear_1")];
            tensor<int32, [4]> var_110 = const()[name = tensor<string, []>("op_110"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_3 = reshape(shape = var_110, x = x_1)[name = tensor<string, []>("x_3")];
            tensor<fp32, [1, 512, 768]> x_5 = linear(bias = text_branch_encoder_layer_0_attention_self_value_bias, weight = text_branch_encoder_layer_0_attention_self_value_weight, x = input_7)[name = tensor<string, []>("linear_2")];
            tensor<int32, [4]> var_119 = const()[name = tensor<string, []>("op_119"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_7 = reshape(shape = var_119, x = x_5)[name = tensor<string, []>("x_7")];
            tensor<int32, [4]> var_121 = const()[name = tensor<string, []>("op_121"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_125 = const()[name = tensor<string, []>("op_125"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_11 = reshape(shape = var_125, x = x_9)[name = tensor<string, []>("x_11")];
            tensor<bool, []> attention_scores_1_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_1_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_37 = transpose(perm = transpose_37_perm_0, x = x_3)[name = tensor<string, []>("transpose_105")];
            tensor<fp32, [1, 12, 512, 64]> transpose_36 = transpose(perm = transpose_36_perm_0, x = x_11)[name = tensor<string, []>("transpose_106")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_1 = matmul(transpose_x = attention_scores_1_transpose_x_0, transpose_y = attention_scores_1_transpose_y_0, x = transpose_36, y = transpose_37)[name = tensor<string, []>("attention_scores_1")];
            tensor<fp32, []> _inversed_attention_scores_3_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_3_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_3 = mul(x = attention_scores_1, y = _inversed_attention_scores_3_y_0)[name = tensor<string, []>("_inversed_attention_scores_3")];
            tensor<fp32, [1, 12, 512, 512]> input_11 = add(x = _inversed_attention_scores_3, y = attention_mask_1)[name = tensor<string, []>("input_11")];
            tensor<fp32, [1, 12, 512, 512]> input_13 = softmax(axis = var_10, x = input_11)[name = tensor<string, []>("input_13")];
            tensor<bool, []> context_layer_1_transpose_x_0 = const()[name = tensor<string, []>("context_layer_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_1_transpose_y_0 = const()[name = tensor<string, []>("context_layer_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_1 = transpose(perm = var_121, x = x_7)[name = tensor<string, []>("transpose_107")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_1 = matmul(transpose_x = context_layer_1_transpose_x_0, transpose_y = context_layer_1_transpose_y_0, x = input_13, y = value_layer_1)[name = tensor<string, []>("context_layer_1")];
            tensor<int32, [4]> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_142 = const()[name = tensor<string, []>("op_142"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_138 = transpose(perm = var_137, x = context_layer_1)[name = tensor<string, []>("transpose_104")];
            tensor<fp32, [1, 512, 768]> input_15 = reshape(shape = var_142, x = var_138)[name = tensor<string, []>("input_15")];
            tensor<fp32, [1, 512, 768]> input_17 = linear(bias = text_branch_encoder_layer_0_attention_output_dense_bias, weight = text_branch_encoder_layer_0_attention_output_dense_weight, x = input_15)[name = tensor<string, []>("linear_3")];
            tensor<fp32, [1, 512, 768]> input_19 = add(x = input_17, y = input_7)[name = tensor<string, []>("input_19")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_21 = layer_norm(axes = input_21_axes_0, beta = text_branch_encoder_layer_0_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_0_attention_output_LayerNorm_weight, x = input_19)[name = tensor<string, []>("input_21")];
            tensor<fp32, [1, 512, 3072]> input_23 = linear(bias = text_branch_encoder_layer_0_intermediate_dense_bias, weight = text_branch_encoder_layer_0_intermediate_dense_weight, x = input_21)[name = tensor<string, []>("linear_4")];
            tensor<string, []> input_25_mode_0 = const()[name = tensor<string, []>("input_25_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_25 = gelu(mode = input_25_mode_0, x = input_23)[name = tensor<string, []>("input_25")];
            tensor<fp32, [1, 512, 768]> input_27 = linear(bias = text_branch_encoder_layer_0_output_dense_bias, weight = text_branch_encoder_layer_0_output_dense_weight, x = input_25)[name = tensor<string, []>("linear_5")];
            tensor<fp32, [1, 512, 768]> input_29 = add(x = input_27, y = input_21)[name = tensor<string, []>("input_29")];
            tensor<int32, [1]> input_31_axes_0 = const()[name = tensor<string, []>("input_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_31 = layer_norm(axes = input_31_axes_0, beta = text_branch_encoder_layer_0_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_0_output_LayerNorm_weight, x = input_29)[name = tensor<string, []>("input_31")];
            tensor<fp32, [1, 512, 768]> x_21 = linear(bias = text_branch_encoder_layer_1_attention_self_query_bias, weight = text_branch_encoder_layer_1_attention_self_query_weight, x = input_31)[name = tensor<string, []>("linear_6")];
            tensor<fp32, [1, 512, 768]> x_13 = linear(bias = text_branch_encoder_layer_1_attention_self_key_bias, weight = text_branch_encoder_layer_1_attention_self_key_weight, x = input_31)[name = tensor<string, []>("linear_7")];
            tensor<int32, [4]> var_187 = const()[name = tensor<string, []>("op_187"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_15 = reshape(shape = var_187, x = x_13)[name = tensor<string, []>("x_15")];
            tensor<fp32, [1, 512, 768]> x_17 = linear(bias = text_branch_encoder_layer_1_attention_self_value_bias, weight = text_branch_encoder_layer_1_attention_self_value_weight, x = input_31)[name = tensor<string, []>("linear_8")];
            tensor<int32, [4]> var_196 = const()[name = tensor<string, []>("op_196"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_19 = reshape(shape = var_196, x = x_17)[name = tensor<string, []>("x_19")];
            tensor<int32, [4]> var_198 = const()[name = tensor<string, []>("op_198"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_202 = const()[name = tensor<string, []>("op_202"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_23 = reshape(shape = var_202, x = x_21)[name = tensor<string, []>("x_23")];
            tensor<bool, []> attention_scores_5_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_5_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_39_perm_0 = const()[name = tensor<string, []>("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_39 = transpose(perm = transpose_39_perm_0, x = x_15)[name = tensor<string, []>("transpose_101")];
            tensor<fp32, [1, 12, 512, 64]> transpose_38 = transpose(perm = transpose_38_perm_0, x = x_23)[name = tensor<string, []>("transpose_102")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_5 = matmul(transpose_x = attention_scores_5_transpose_x_0, transpose_y = attention_scores_5_transpose_y_0, x = transpose_38, y = transpose_39)[name = tensor<string, []>("attention_scores_5")];
            tensor<fp32, []> _inversed_attention_scores_7_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_7_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_7 = mul(x = attention_scores_5, y = _inversed_attention_scores_7_y_0)[name = tensor<string, []>("_inversed_attention_scores_7")];
            tensor<fp32, [1, 12, 512, 512]> input_33 = add(x = _inversed_attention_scores_7, y = attention_mask_1)[name = tensor<string, []>("input_33")];
            tensor<fp32, [1, 12, 512, 512]> input_35 = softmax(axis = var_10, x = input_33)[name = tensor<string, []>("input_35")];
            tensor<bool, []> context_layer_5_transpose_x_0 = const()[name = tensor<string, []>("context_layer_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_5_transpose_y_0 = const()[name = tensor<string, []>("context_layer_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_3 = transpose(perm = var_198, x = x_19)[name = tensor<string, []>("transpose_103")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_5 = matmul(transpose_x = context_layer_5_transpose_x_0, transpose_y = context_layer_5_transpose_y_0, x = input_35, y = value_layer_3)[name = tensor<string, []>("context_layer_5")];
            tensor<int32, [4]> var_214 = const()[name = tensor<string, []>("op_214"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_219 = const()[name = tensor<string, []>("op_219"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_215 = transpose(perm = var_214, x = context_layer_5)[name = tensor<string, []>("transpose_100")];
            tensor<fp32, [1, 512, 768]> input_37 = reshape(shape = var_219, x = var_215)[name = tensor<string, []>("input_37")];
            tensor<fp32, [1, 512, 768]> input_39 = linear(bias = text_branch_encoder_layer_1_attention_output_dense_bias, weight = text_branch_encoder_layer_1_attention_output_dense_weight, x = input_37)[name = tensor<string, []>("linear_9")];
            tensor<fp32, [1, 512, 768]> input_41 = add(x = input_39, y = input_31)[name = tensor<string, []>("input_41")];
            tensor<int32, [1]> input_43_axes_0 = const()[name = tensor<string, []>("input_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_43 = layer_norm(axes = input_43_axes_0, beta = text_branch_encoder_layer_1_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_1_attention_output_LayerNorm_weight, x = input_41)[name = tensor<string, []>("input_43")];
            tensor<fp32, [1, 512, 3072]> input_45 = linear(bias = text_branch_encoder_layer_1_intermediate_dense_bias, weight = text_branch_encoder_layer_1_intermediate_dense_weight, x = input_43)[name = tensor<string, []>("linear_10")];
            tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_47 = gelu(mode = input_47_mode_0, x = input_45)[name = tensor<string, []>("input_47")];
            tensor<fp32, [1, 512, 768]> input_49 = linear(bias = text_branch_encoder_layer_1_output_dense_bias, weight = text_branch_encoder_layer_1_output_dense_weight, x = input_47)[name = tensor<string, []>("linear_11")];
            tensor<fp32, [1, 512, 768]> input_51 = add(x = input_49, y = input_43)[name = tensor<string, []>("input_51")];
            tensor<int32, [1]> input_53_axes_0 = const()[name = tensor<string, []>("input_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_53 = layer_norm(axes = input_53_axes_0, beta = text_branch_encoder_layer_1_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_1_output_LayerNorm_weight, x = input_51)[name = tensor<string, []>("input_53")];
            tensor<fp32, [1, 512, 768]> x_33 = linear(bias = text_branch_encoder_layer_2_attention_self_query_bias, weight = text_branch_encoder_layer_2_attention_self_query_weight, x = input_53)[name = tensor<string, []>("linear_12")];
            tensor<fp32, [1, 512, 768]> x_25 = linear(bias = text_branch_encoder_layer_2_attention_self_key_bias, weight = text_branch_encoder_layer_2_attention_self_key_weight, x = input_53)[name = tensor<string, []>("linear_13")];
            tensor<int32, [4]> var_264 = const()[name = tensor<string, []>("op_264"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_27 = reshape(shape = var_264, x = x_25)[name = tensor<string, []>("x_27")];
            tensor<fp32, [1, 512, 768]> x_29 = linear(bias = text_branch_encoder_layer_2_attention_self_value_bias, weight = text_branch_encoder_layer_2_attention_self_value_weight, x = input_53)[name = tensor<string, []>("linear_14")];
            tensor<int32, [4]> var_273 = const()[name = tensor<string, []>("op_273"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_31 = reshape(shape = var_273, x = x_29)[name = tensor<string, []>("x_31")];
            tensor<int32, [4]> var_275 = const()[name = tensor<string, []>("op_275"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_279 = const()[name = tensor<string, []>("op_279"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_35 = reshape(shape = var_279, x = x_33)[name = tensor<string, []>("x_35")];
            tensor<bool, []> attention_scores_9_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_9_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_40_perm_0 = const()[name = tensor<string, []>("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_41_perm_0 = const()[name = tensor<string, []>("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_41 = transpose(perm = transpose_41_perm_0, x = x_27)[name = tensor<string, []>("transpose_97")];
            tensor<fp32, [1, 12, 512, 64]> transpose_40 = transpose(perm = transpose_40_perm_0, x = x_35)[name = tensor<string, []>("transpose_98")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_9 = matmul(transpose_x = attention_scores_9_transpose_x_0, transpose_y = attention_scores_9_transpose_y_0, x = transpose_40, y = transpose_41)[name = tensor<string, []>("attention_scores_9")];
            tensor<fp32, []> _inversed_attention_scores_11_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_11_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_11 = mul(x = attention_scores_9, y = _inversed_attention_scores_11_y_0)[name = tensor<string, []>("_inversed_attention_scores_11")];
            tensor<fp32, [1, 12, 512, 512]> input_55 = add(x = _inversed_attention_scores_11, y = attention_mask_1)[name = tensor<string, []>("input_55")];
            tensor<fp32, [1, 12, 512, 512]> input_57 = softmax(axis = var_10, x = input_55)[name = tensor<string, []>("input_57")];
            tensor<bool, []> context_layer_9_transpose_x_0 = const()[name = tensor<string, []>("context_layer_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_9_transpose_y_0 = const()[name = tensor<string, []>("context_layer_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_5 = transpose(perm = var_275, x = x_31)[name = tensor<string, []>("transpose_99")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_9 = matmul(transpose_x = context_layer_9_transpose_x_0, transpose_y = context_layer_9_transpose_y_0, x = input_57, y = value_layer_5)[name = tensor<string, []>("context_layer_9")];
            tensor<int32, [4]> var_291 = const()[name = tensor<string, []>("op_291"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_296 = const()[name = tensor<string, []>("op_296"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_292 = transpose(perm = var_291, x = context_layer_9)[name = tensor<string, []>("transpose_96")];
            tensor<fp32, [1, 512, 768]> input_59 = reshape(shape = var_296, x = var_292)[name = tensor<string, []>("input_59")];
            tensor<fp32, [1, 512, 768]> input_61 = linear(bias = text_branch_encoder_layer_2_attention_output_dense_bias, weight = text_branch_encoder_layer_2_attention_output_dense_weight, x = input_59)[name = tensor<string, []>("linear_15")];
            tensor<fp32, [1, 512, 768]> input_63 = add(x = input_61, y = input_53)[name = tensor<string, []>("input_63")];
            tensor<int32, [1]> input_65_axes_0 = const()[name = tensor<string, []>("input_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_65 = layer_norm(axes = input_65_axes_0, beta = text_branch_encoder_layer_2_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_2_attention_output_LayerNorm_weight, x = input_63)[name = tensor<string, []>("input_65")];
            tensor<fp32, [1, 512, 3072]> input_67 = linear(bias = text_branch_encoder_layer_2_intermediate_dense_bias, weight = text_branch_encoder_layer_2_intermediate_dense_weight, x = input_65)[name = tensor<string, []>("linear_16")];
            tensor<string, []> input_69_mode_0 = const()[name = tensor<string, []>("input_69_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_69 = gelu(mode = input_69_mode_0, x = input_67)[name = tensor<string, []>("input_69")];
            tensor<fp32, [1, 512, 768]> input_71 = linear(bias = text_branch_encoder_layer_2_output_dense_bias, weight = text_branch_encoder_layer_2_output_dense_weight, x = input_69)[name = tensor<string, []>("linear_17")];
            tensor<fp32, [1, 512, 768]> input_73 = add(x = input_71, y = input_65)[name = tensor<string, []>("input_73")];
            tensor<int32, [1]> input_75_axes_0 = const()[name = tensor<string, []>("input_75_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_75 = layer_norm(axes = input_75_axes_0, beta = text_branch_encoder_layer_2_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_2_output_LayerNorm_weight, x = input_73)[name = tensor<string, []>("input_75")];
            tensor<fp32, [1, 512, 768]> x_45 = linear(bias = text_branch_encoder_layer_3_attention_self_query_bias, weight = text_branch_encoder_layer_3_attention_self_query_weight, x = input_75)[name = tensor<string, []>("linear_18")];
            tensor<fp32, [1, 512, 768]> x_37 = linear(bias = text_branch_encoder_layer_3_attention_self_key_bias, weight = text_branch_encoder_layer_3_attention_self_key_weight, x = input_75)[name = tensor<string, []>("linear_19")];
            tensor<int32, [4]> var_341 = const()[name = tensor<string, []>("op_341"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_39 = reshape(shape = var_341, x = x_37)[name = tensor<string, []>("x_39")];
            tensor<fp32, [1, 512, 768]> x_41 = linear(bias = text_branch_encoder_layer_3_attention_self_value_bias, weight = text_branch_encoder_layer_3_attention_self_value_weight, x = input_75)[name = tensor<string, []>("linear_20")];
            tensor<int32, [4]> var_350 = const()[name = tensor<string, []>("op_350"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_43 = reshape(shape = var_350, x = x_41)[name = tensor<string, []>("x_43")];
            tensor<int32, [4]> var_352 = const()[name = tensor<string, []>("op_352"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_356 = const()[name = tensor<string, []>("op_356"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_47 = reshape(shape = var_356, x = x_45)[name = tensor<string, []>("x_47")];
            tensor<bool, []> attention_scores_13_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_13_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_42_perm_0 = const()[name = tensor<string, []>("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_43_perm_0 = const()[name = tensor<string, []>("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_43 = transpose(perm = transpose_43_perm_0, x = x_39)[name = tensor<string, []>("transpose_93")];
            tensor<fp32, [1, 12, 512, 64]> transpose_42 = transpose(perm = transpose_42_perm_0, x = x_47)[name = tensor<string, []>("transpose_94")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_13 = matmul(transpose_x = attention_scores_13_transpose_x_0, transpose_y = attention_scores_13_transpose_y_0, x = transpose_42, y = transpose_43)[name = tensor<string, []>("attention_scores_13")];
            tensor<fp32, []> _inversed_attention_scores_15_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_15_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_15 = mul(x = attention_scores_13, y = _inversed_attention_scores_15_y_0)[name = tensor<string, []>("_inversed_attention_scores_15")];
            tensor<fp32, [1, 12, 512, 512]> input_77 = add(x = _inversed_attention_scores_15, y = attention_mask_1)[name = tensor<string, []>("input_77")];
            tensor<fp32, [1, 12, 512, 512]> input_79 = softmax(axis = var_10, x = input_77)[name = tensor<string, []>("input_79")];
            tensor<bool, []> context_layer_13_transpose_x_0 = const()[name = tensor<string, []>("context_layer_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_13_transpose_y_0 = const()[name = tensor<string, []>("context_layer_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_7 = transpose(perm = var_352, x = x_43)[name = tensor<string, []>("transpose_95")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_13 = matmul(transpose_x = context_layer_13_transpose_x_0, transpose_y = context_layer_13_transpose_y_0, x = input_79, y = value_layer_7)[name = tensor<string, []>("context_layer_13")];
            tensor<int32, [4]> var_368 = const()[name = tensor<string, []>("op_368"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_373 = const()[name = tensor<string, []>("op_373"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_369 = transpose(perm = var_368, x = context_layer_13)[name = tensor<string, []>("transpose_92")];
            tensor<fp32, [1, 512, 768]> input_81 = reshape(shape = var_373, x = var_369)[name = tensor<string, []>("input_81")];
            tensor<fp32, [1, 512, 768]> input_83 = linear(bias = text_branch_encoder_layer_3_attention_output_dense_bias, weight = text_branch_encoder_layer_3_attention_output_dense_weight, x = input_81)[name = tensor<string, []>("linear_21")];
            tensor<fp32, [1, 512, 768]> input_85 = add(x = input_83, y = input_75)[name = tensor<string, []>("input_85")];
            tensor<int32, [1]> input_87_axes_0 = const()[name = tensor<string, []>("input_87_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_87 = layer_norm(axes = input_87_axes_0, beta = text_branch_encoder_layer_3_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_3_attention_output_LayerNorm_weight, x = input_85)[name = tensor<string, []>("input_87")];
            tensor<fp32, [1, 512, 3072]> input_89 = linear(bias = text_branch_encoder_layer_3_intermediate_dense_bias, weight = text_branch_encoder_layer_3_intermediate_dense_weight, x = input_87)[name = tensor<string, []>("linear_22")];
            tensor<string, []> input_91_mode_0 = const()[name = tensor<string, []>("input_91_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_91 = gelu(mode = input_91_mode_0, x = input_89)[name = tensor<string, []>("input_91")];
            tensor<fp32, [1, 512, 768]> input_93 = linear(bias = text_branch_encoder_layer_3_output_dense_bias, weight = text_branch_encoder_layer_3_output_dense_weight, x = input_91)[name = tensor<string, []>("linear_23")];
            tensor<fp32, [1, 512, 768]> input_95 = add(x = input_93, y = input_87)[name = tensor<string, []>("input_95")];
            tensor<int32, [1]> input_97_axes_0 = const()[name = tensor<string, []>("input_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_97 = layer_norm(axes = input_97_axes_0, beta = text_branch_encoder_layer_3_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_3_output_LayerNorm_weight, x = input_95)[name = tensor<string, []>("input_97")];
            tensor<fp32, [1, 512, 768]> x_57 = linear(bias = text_branch_encoder_layer_4_attention_self_query_bias, weight = text_branch_encoder_layer_4_attention_self_query_weight, x = input_97)[name = tensor<string, []>("linear_24")];
            tensor<fp32, [1, 512, 768]> x_49 = linear(bias = text_branch_encoder_layer_4_attention_self_key_bias, weight = text_branch_encoder_layer_4_attention_self_key_weight, x = input_97)[name = tensor<string, []>("linear_25")];
            tensor<int32, [4]> var_418 = const()[name = tensor<string, []>("op_418"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_51 = reshape(shape = var_418, x = x_49)[name = tensor<string, []>("x_51")];
            tensor<fp32, [1, 512, 768]> x_53 = linear(bias = text_branch_encoder_layer_4_attention_self_value_bias, weight = text_branch_encoder_layer_4_attention_self_value_weight, x = input_97)[name = tensor<string, []>("linear_26")];
            tensor<int32, [4]> var_427 = const()[name = tensor<string, []>("op_427"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_55 = reshape(shape = var_427, x = x_53)[name = tensor<string, []>("x_55")];
            tensor<int32, [4]> var_429 = const()[name = tensor<string, []>("op_429"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_433 = const()[name = tensor<string, []>("op_433"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_59 = reshape(shape = var_433, x = x_57)[name = tensor<string, []>("x_59")];
            tensor<bool, []> attention_scores_17_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_17_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_44_perm_0 = const()[name = tensor<string, []>("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_45_perm_0 = const()[name = tensor<string, []>("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_45 = transpose(perm = transpose_45_perm_0, x = x_51)[name = tensor<string, []>("transpose_89")];
            tensor<fp32, [1, 12, 512, 64]> transpose_44 = transpose(perm = transpose_44_perm_0, x = x_59)[name = tensor<string, []>("transpose_90")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_17 = matmul(transpose_x = attention_scores_17_transpose_x_0, transpose_y = attention_scores_17_transpose_y_0, x = transpose_44, y = transpose_45)[name = tensor<string, []>("attention_scores_17")];
            tensor<fp32, []> _inversed_attention_scores_19_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_19_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_19 = mul(x = attention_scores_17, y = _inversed_attention_scores_19_y_0)[name = tensor<string, []>("_inversed_attention_scores_19")];
            tensor<fp32, [1, 12, 512, 512]> input_99 = add(x = _inversed_attention_scores_19, y = attention_mask_1)[name = tensor<string, []>("input_99")];
            tensor<fp32, [1, 12, 512, 512]> input_101 = softmax(axis = var_10, x = input_99)[name = tensor<string, []>("input_101")];
            tensor<bool, []> context_layer_17_transpose_x_0 = const()[name = tensor<string, []>("context_layer_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_17_transpose_y_0 = const()[name = tensor<string, []>("context_layer_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_9 = transpose(perm = var_429, x = x_55)[name = tensor<string, []>("transpose_91")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_17 = matmul(transpose_x = context_layer_17_transpose_x_0, transpose_y = context_layer_17_transpose_y_0, x = input_101, y = value_layer_9)[name = tensor<string, []>("context_layer_17")];
            tensor<int32, [4]> var_445 = const()[name = tensor<string, []>("op_445"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_450 = const()[name = tensor<string, []>("op_450"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_446 = transpose(perm = var_445, x = context_layer_17)[name = tensor<string, []>("transpose_88")];
            tensor<fp32, [1, 512, 768]> input_103 = reshape(shape = var_450, x = var_446)[name = tensor<string, []>("input_103")];
            tensor<fp32, [1, 512, 768]> input_105 = linear(bias = text_branch_encoder_layer_4_attention_output_dense_bias, weight = text_branch_encoder_layer_4_attention_output_dense_weight, x = input_103)[name = tensor<string, []>("linear_27")];
            tensor<fp32, [1, 512, 768]> input_107 = add(x = input_105, y = input_97)[name = tensor<string, []>("input_107")];
            tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_109 = layer_norm(axes = input_109_axes_0, beta = text_branch_encoder_layer_4_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_4_attention_output_LayerNorm_weight, x = input_107)[name = tensor<string, []>("input_109")];
            tensor<fp32, [1, 512, 3072]> input_111 = linear(bias = text_branch_encoder_layer_4_intermediate_dense_bias, weight = text_branch_encoder_layer_4_intermediate_dense_weight, x = input_109)[name = tensor<string, []>("linear_28")];
            tensor<string, []> input_113_mode_0 = const()[name = tensor<string, []>("input_113_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_113 = gelu(mode = input_113_mode_0, x = input_111)[name = tensor<string, []>("input_113")];
            tensor<fp32, [1, 512, 768]> input_115 = linear(bias = text_branch_encoder_layer_4_output_dense_bias, weight = text_branch_encoder_layer_4_output_dense_weight, x = input_113)[name = tensor<string, []>("linear_29")];
            tensor<fp32, [1, 512, 768]> input_117 = add(x = input_115, y = input_109)[name = tensor<string, []>("input_117")];
            tensor<int32, [1]> input_119_axes_0 = const()[name = tensor<string, []>("input_119_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_119 = layer_norm(axes = input_119_axes_0, beta = text_branch_encoder_layer_4_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_4_output_LayerNorm_weight, x = input_117)[name = tensor<string, []>("input_119")];
            tensor<fp32, [1, 512, 768]> x_69 = linear(bias = text_branch_encoder_layer_5_attention_self_query_bias, weight = text_branch_encoder_layer_5_attention_self_query_weight, x = input_119)[name = tensor<string, []>("linear_30")];
            tensor<fp32, [1, 512, 768]> x_61 = linear(bias = text_branch_encoder_layer_5_attention_self_key_bias, weight = text_branch_encoder_layer_5_attention_self_key_weight, x = input_119)[name = tensor<string, []>("linear_31")];
            tensor<int32, [4]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_63 = reshape(shape = var_495, x = x_61)[name = tensor<string, []>("x_63")];
            tensor<fp32, [1, 512, 768]> x_65 = linear(bias = text_branch_encoder_layer_5_attention_self_value_bias, weight = text_branch_encoder_layer_5_attention_self_value_weight, x = input_119)[name = tensor<string, []>("linear_32")];
            tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_67 = reshape(shape = var_504, x = x_65)[name = tensor<string, []>("x_67")];
            tensor<int32, [4]> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_510 = const()[name = tensor<string, []>("op_510"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_71 = reshape(shape = var_510, x = x_69)[name = tensor<string, []>("x_71")];
            tensor<bool, []> attention_scores_21_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_21_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_46_perm_0 = const()[name = tensor<string, []>("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_47_perm_0 = const()[name = tensor<string, []>("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_47 = transpose(perm = transpose_47_perm_0, x = x_63)[name = tensor<string, []>("transpose_85")];
            tensor<fp32, [1, 12, 512, 64]> transpose_46 = transpose(perm = transpose_46_perm_0, x = x_71)[name = tensor<string, []>("transpose_86")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_21 = matmul(transpose_x = attention_scores_21_transpose_x_0, transpose_y = attention_scores_21_transpose_y_0, x = transpose_46, y = transpose_47)[name = tensor<string, []>("attention_scores_21")];
            tensor<fp32, []> _inversed_attention_scores_23_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_23_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_23 = mul(x = attention_scores_21, y = _inversed_attention_scores_23_y_0)[name = tensor<string, []>("_inversed_attention_scores_23")];
            tensor<fp32, [1, 12, 512, 512]> input_121 = add(x = _inversed_attention_scores_23, y = attention_mask_1)[name = tensor<string, []>("input_121")];
            tensor<fp32, [1, 12, 512, 512]> input_123 = softmax(axis = var_10, x = input_121)[name = tensor<string, []>("input_123")];
            tensor<bool, []> context_layer_21_transpose_x_0 = const()[name = tensor<string, []>("context_layer_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_21_transpose_y_0 = const()[name = tensor<string, []>("context_layer_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_11 = transpose(perm = var_506, x = x_67)[name = tensor<string, []>("transpose_87")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_21 = matmul(transpose_x = context_layer_21_transpose_x_0, transpose_y = context_layer_21_transpose_y_0, x = input_123, y = value_layer_11)[name = tensor<string, []>("context_layer_21")];
            tensor<int32, [4]> var_522 = const()[name = tensor<string, []>("op_522"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_527 = const()[name = tensor<string, []>("op_527"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_523 = transpose(perm = var_522, x = context_layer_21)[name = tensor<string, []>("transpose_84")];
            tensor<fp32, [1, 512, 768]> input_125 = reshape(shape = var_527, x = var_523)[name = tensor<string, []>("input_125")];
            tensor<fp32, [1, 512, 768]> input_127 = linear(bias = text_branch_encoder_layer_5_attention_output_dense_bias, weight = text_branch_encoder_layer_5_attention_output_dense_weight, x = input_125)[name = tensor<string, []>("linear_33")];
            tensor<fp32, [1, 512, 768]> input_129 = add(x = input_127, y = input_119)[name = tensor<string, []>("input_129")];
            tensor<int32, [1]> input_131_axes_0 = const()[name = tensor<string, []>("input_131_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_131 = layer_norm(axes = input_131_axes_0, beta = text_branch_encoder_layer_5_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_5_attention_output_LayerNorm_weight, x = input_129)[name = tensor<string, []>("input_131")];
            tensor<fp32, [1, 512, 3072]> input_133 = linear(bias = text_branch_encoder_layer_5_intermediate_dense_bias, weight = text_branch_encoder_layer_5_intermediate_dense_weight, x = input_131)[name = tensor<string, []>("linear_34")];
            tensor<string, []> input_135_mode_0 = const()[name = tensor<string, []>("input_135_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_135 = gelu(mode = input_135_mode_0, x = input_133)[name = tensor<string, []>("input_135")];
            tensor<fp32, [1, 512, 768]> input_137 = linear(bias = text_branch_encoder_layer_5_output_dense_bias, weight = text_branch_encoder_layer_5_output_dense_weight, x = input_135)[name = tensor<string, []>("linear_35")];
            tensor<fp32, [1, 512, 768]> input_139 = add(x = input_137, y = input_131)[name = tensor<string, []>("input_139")];
            tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_141 = layer_norm(axes = input_141_axes_0, beta = text_branch_encoder_layer_5_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_5_output_LayerNorm_weight, x = input_139)[name = tensor<string, []>("input_141")];
            tensor<fp32, [1, 512, 768]> x_81 = linear(bias = text_branch_encoder_layer_6_attention_self_query_bias, weight = text_branch_encoder_layer_6_attention_self_query_weight, x = input_141)[name = tensor<string, []>("linear_36")];
            tensor<fp32, [1, 512, 768]> x_73 = linear(bias = text_branch_encoder_layer_6_attention_self_key_bias, weight = text_branch_encoder_layer_6_attention_self_key_weight, x = input_141)[name = tensor<string, []>("linear_37")];
            tensor<int32, [4]> var_572 = const()[name = tensor<string, []>("op_572"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_75 = reshape(shape = var_572, x = x_73)[name = tensor<string, []>("x_75")];
            tensor<fp32, [1, 512, 768]> x_77 = linear(bias = text_branch_encoder_layer_6_attention_self_value_bias, weight = text_branch_encoder_layer_6_attention_self_value_weight, x = input_141)[name = tensor<string, []>("linear_38")];
            tensor<int32, [4]> var_581 = const()[name = tensor<string, []>("op_581"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_79 = reshape(shape = var_581, x = x_77)[name = tensor<string, []>("x_79")];
            tensor<int32, [4]> var_583 = const()[name = tensor<string, []>("op_583"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_587 = const()[name = tensor<string, []>("op_587"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_83 = reshape(shape = var_587, x = x_81)[name = tensor<string, []>("x_83")];
            tensor<bool, []> attention_scores_25_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_25_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = tensor<string, []>("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_49_perm_0 = const()[name = tensor<string, []>("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_49 = transpose(perm = transpose_49_perm_0, x = x_75)[name = tensor<string, []>("transpose_81")];
            tensor<fp32, [1, 12, 512, 64]> transpose_48 = transpose(perm = transpose_48_perm_0, x = x_83)[name = tensor<string, []>("transpose_82")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_25 = matmul(transpose_x = attention_scores_25_transpose_x_0, transpose_y = attention_scores_25_transpose_y_0, x = transpose_48, y = transpose_49)[name = tensor<string, []>("attention_scores_25")];
            tensor<fp32, []> _inversed_attention_scores_27_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_27_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_27 = mul(x = attention_scores_25, y = _inversed_attention_scores_27_y_0)[name = tensor<string, []>("_inversed_attention_scores_27")];
            tensor<fp32, [1, 12, 512, 512]> input_143 = add(x = _inversed_attention_scores_27, y = attention_mask_1)[name = tensor<string, []>("input_143")];
            tensor<fp32, [1, 12, 512, 512]> input_145 = softmax(axis = var_10, x = input_143)[name = tensor<string, []>("input_145")];
            tensor<bool, []> context_layer_25_transpose_x_0 = const()[name = tensor<string, []>("context_layer_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_25_transpose_y_0 = const()[name = tensor<string, []>("context_layer_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_13 = transpose(perm = var_583, x = x_79)[name = tensor<string, []>("transpose_83")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_25 = matmul(transpose_x = context_layer_25_transpose_x_0, transpose_y = context_layer_25_transpose_y_0, x = input_145, y = value_layer_13)[name = tensor<string, []>("context_layer_25")];
            tensor<int32, [4]> var_599 = const()[name = tensor<string, []>("op_599"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_604 = const()[name = tensor<string, []>("op_604"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_600 = transpose(perm = var_599, x = context_layer_25)[name = tensor<string, []>("transpose_80")];
            tensor<fp32, [1, 512, 768]> input_147 = reshape(shape = var_604, x = var_600)[name = tensor<string, []>("input_147")];
            tensor<fp32, [1, 512, 768]> input_149 = linear(bias = text_branch_encoder_layer_6_attention_output_dense_bias, weight = text_branch_encoder_layer_6_attention_output_dense_weight, x = input_147)[name = tensor<string, []>("linear_39")];
            tensor<fp32, [1, 512, 768]> input_151 = add(x = input_149, y = input_141)[name = tensor<string, []>("input_151")];
            tensor<int32, [1]> input_153_axes_0 = const()[name = tensor<string, []>("input_153_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_153 = layer_norm(axes = input_153_axes_0, beta = text_branch_encoder_layer_6_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_6_attention_output_LayerNorm_weight, x = input_151)[name = tensor<string, []>("input_153")];
            tensor<fp32, [1, 512, 3072]> input_155 = linear(bias = text_branch_encoder_layer_6_intermediate_dense_bias, weight = text_branch_encoder_layer_6_intermediate_dense_weight, x = input_153)[name = tensor<string, []>("linear_40")];
            tensor<string, []> input_157_mode_0 = const()[name = tensor<string, []>("input_157_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_157 = gelu(mode = input_157_mode_0, x = input_155)[name = tensor<string, []>("input_157")];
            tensor<fp32, [1, 512, 768]> input_159 = linear(bias = text_branch_encoder_layer_6_output_dense_bias, weight = text_branch_encoder_layer_6_output_dense_weight, x = input_157)[name = tensor<string, []>("linear_41")];
            tensor<fp32, [1, 512, 768]> input_161 = add(x = input_159, y = input_153)[name = tensor<string, []>("input_161")];
            tensor<int32, [1]> input_163_axes_0 = const()[name = tensor<string, []>("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_163 = layer_norm(axes = input_163_axes_0, beta = text_branch_encoder_layer_6_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_6_output_LayerNorm_weight, x = input_161)[name = tensor<string, []>("input_163")];
            tensor<fp32, [1, 512, 768]> x_93 = linear(bias = text_branch_encoder_layer_7_attention_self_query_bias, weight = text_branch_encoder_layer_7_attention_self_query_weight, x = input_163)[name = tensor<string, []>("linear_42")];
            tensor<fp32, [1, 512, 768]> x_85 = linear(bias = text_branch_encoder_layer_7_attention_self_key_bias, weight = text_branch_encoder_layer_7_attention_self_key_weight, x = input_163)[name = tensor<string, []>("linear_43")];
            tensor<int32, [4]> var_649 = const()[name = tensor<string, []>("op_649"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_87 = reshape(shape = var_649, x = x_85)[name = tensor<string, []>("x_87")];
            tensor<fp32, [1, 512, 768]> x_89 = linear(bias = text_branch_encoder_layer_7_attention_self_value_bias, weight = text_branch_encoder_layer_7_attention_self_value_weight, x = input_163)[name = tensor<string, []>("linear_44")];
            tensor<int32, [4]> var_658 = const()[name = tensor<string, []>("op_658"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_91 = reshape(shape = var_658, x = x_89)[name = tensor<string, []>("x_91")];
            tensor<int32, [4]> var_660 = const()[name = tensor<string, []>("op_660"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_664 = const()[name = tensor<string, []>("op_664"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_95 = reshape(shape = var_664, x = x_93)[name = tensor<string, []>("x_95")];
            tensor<bool, []> attention_scores_29_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_29_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_50_perm_0 = const()[name = tensor<string, []>("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_51 = transpose(perm = transpose_51_perm_0, x = x_87)[name = tensor<string, []>("transpose_77")];
            tensor<fp32, [1, 12, 512, 64]> transpose_50 = transpose(perm = transpose_50_perm_0, x = x_95)[name = tensor<string, []>("transpose_78")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_29 = matmul(transpose_x = attention_scores_29_transpose_x_0, transpose_y = attention_scores_29_transpose_y_0, x = transpose_50, y = transpose_51)[name = tensor<string, []>("attention_scores_29")];
            tensor<fp32, []> _inversed_attention_scores_31_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_31_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_31 = mul(x = attention_scores_29, y = _inversed_attention_scores_31_y_0)[name = tensor<string, []>("_inversed_attention_scores_31")];
            tensor<fp32, [1, 12, 512, 512]> input_165 = add(x = _inversed_attention_scores_31, y = attention_mask_1)[name = tensor<string, []>("input_165")];
            tensor<fp32, [1, 12, 512, 512]> input_167 = softmax(axis = var_10, x = input_165)[name = tensor<string, []>("input_167")];
            tensor<bool, []> context_layer_29_transpose_x_0 = const()[name = tensor<string, []>("context_layer_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_29_transpose_y_0 = const()[name = tensor<string, []>("context_layer_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_15 = transpose(perm = var_660, x = x_91)[name = tensor<string, []>("transpose_79")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_29 = matmul(transpose_x = context_layer_29_transpose_x_0, transpose_y = context_layer_29_transpose_y_0, x = input_167, y = value_layer_15)[name = tensor<string, []>("context_layer_29")];
            tensor<int32, [4]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_681 = const()[name = tensor<string, []>("op_681"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_677 = transpose(perm = var_676, x = context_layer_29)[name = tensor<string, []>("transpose_76")];
            tensor<fp32, [1, 512, 768]> input_169 = reshape(shape = var_681, x = var_677)[name = tensor<string, []>("input_169")];
            tensor<fp32, [1, 512, 768]> input_171 = linear(bias = text_branch_encoder_layer_7_attention_output_dense_bias, weight = text_branch_encoder_layer_7_attention_output_dense_weight, x = input_169)[name = tensor<string, []>("linear_45")];
            tensor<fp32, [1, 512, 768]> input_173 = add(x = input_171, y = input_163)[name = tensor<string, []>("input_173")];
            tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_175 = layer_norm(axes = input_175_axes_0, beta = text_branch_encoder_layer_7_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_7_attention_output_LayerNorm_weight, x = input_173)[name = tensor<string, []>("input_175")];
            tensor<fp32, [1, 512, 3072]> input_177 = linear(bias = text_branch_encoder_layer_7_intermediate_dense_bias, weight = text_branch_encoder_layer_7_intermediate_dense_weight, x = input_175)[name = tensor<string, []>("linear_46")];
            tensor<string, []> input_179_mode_0 = const()[name = tensor<string, []>("input_179_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_179 = gelu(mode = input_179_mode_0, x = input_177)[name = tensor<string, []>("input_179")];
            tensor<fp32, [1, 512, 768]> input_181 = linear(bias = text_branch_encoder_layer_7_output_dense_bias, weight = text_branch_encoder_layer_7_output_dense_weight, x = input_179)[name = tensor<string, []>("linear_47")];
            tensor<fp32, [1, 512, 768]> input_183 = add(x = input_181, y = input_175)[name = tensor<string, []>("input_183")];
            tensor<int32, [1]> input_185_axes_0 = const()[name = tensor<string, []>("input_185_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_185 = layer_norm(axes = input_185_axes_0, beta = text_branch_encoder_layer_7_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_7_output_LayerNorm_weight, x = input_183)[name = tensor<string, []>("input_185")];
            tensor<fp32, [1, 512, 768]> x_105 = linear(bias = text_branch_encoder_layer_8_attention_self_query_bias, weight = text_branch_encoder_layer_8_attention_self_query_weight, x = input_185)[name = tensor<string, []>("linear_48")];
            tensor<fp32, [1, 512, 768]> x_97 = linear(bias = text_branch_encoder_layer_8_attention_self_key_bias, weight = text_branch_encoder_layer_8_attention_self_key_weight, x = input_185)[name = tensor<string, []>("linear_49")];
            tensor<int32, [4]> var_726 = const()[name = tensor<string, []>("op_726"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_99 = reshape(shape = var_726, x = x_97)[name = tensor<string, []>("x_99")];
            tensor<fp32, [1, 512, 768]> x_101 = linear(bias = text_branch_encoder_layer_8_attention_self_value_bias, weight = text_branch_encoder_layer_8_attention_self_value_weight, x = input_185)[name = tensor<string, []>("linear_50")];
            tensor<int32, [4]> var_735 = const()[name = tensor<string, []>("op_735"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_103 = reshape(shape = var_735, x = x_101)[name = tensor<string, []>("x_103")];
            tensor<int32, [4]> var_737 = const()[name = tensor<string, []>("op_737"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_741 = const()[name = tensor<string, []>("op_741"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_107 = reshape(shape = var_741, x = x_105)[name = tensor<string, []>("x_107")];
            tensor<bool, []> attention_scores_33_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_33_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = tensor<string, []>("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = tensor<string, []>("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_53 = transpose(perm = transpose_53_perm_0, x = x_99)[name = tensor<string, []>("transpose_73")];
            tensor<fp32, [1, 12, 512, 64]> transpose_52 = transpose(perm = transpose_52_perm_0, x = x_107)[name = tensor<string, []>("transpose_74")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_33 = matmul(transpose_x = attention_scores_33_transpose_x_0, transpose_y = attention_scores_33_transpose_y_0, x = transpose_52, y = transpose_53)[name = tensor<string, []>("attention_scores_33")];
            tensor<fp32, []> _inversed_attention_scores_35_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_35_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_35 = mul(x = attention_scores_33, y = _inversed_attention_scores_35_y_0)[name = tensor<string, []>("_inversed_attention_scores_35")];
            tensor<fp32, [1, 12, 512, 512]> input_187 = add(x = _inversed_attention_scores_35, y = attention_mask_1)[name = tensor<string, []>("input_187")];
            tensor<fp32, [1, 12, 512, 512]> input_189 = softmax(axis = var_10, x = input_187)[name = tensor<string, []>("input_189")];
            tensor<bool, []> context_layer_33_transpose_x_0 = const()[name = tensor<string, []>("context_layer_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_33_transpose_y_0 = const()[name = tensor<string, []>("context_layer_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_17 = transpose(perm = var_737, x = x_103)[name = tensor<string, []>("transpose_75")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_33 = matmul(transpose_x = context_layer_33_transpose_x_0, transpose_y = context_layer_33_transpose_y_0, x = input_189, y = value_layer_17)[name = tensor<string, []>("context_layer_33")];
            tensor<int32, [4]> var_753 = const()[name = tensor<string, []>("op_753"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_754 = transpose(perm = var_753, x = context_layer_33)[name = tensor<string, []>("transpose_72")];
            tensor<fp32, [1, 512, 768]> input_191 = reshape(shape = var_758, x = var_754)[name = tensor<string, []>("input_191")];
            tensor<fp32, [1, 512, 768]> input_193 = linear(bias = text_branch_encoder_layer_8_attention_output_dense_bias, weight = text_branch_encoder_layer_8_attention_output_dense_weight, x = input_191)[name = tensor<string, []>("linear_51")];
            tensor<fp32, [1, 512, 768]> input_195 = add(x = input_193, y = input_185)[name = tensor<string, []>("input_195")];
            tensor<int32, [1]> input_197_axes_0 = const()[name = tensor<string, []>("input_197_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_197 = layer_norm(axes = input_197_axes_0, beta = text_branch_encoder_layer_8_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_8_attention_output_LayerNorm_weight, x = input_195)[name = tensor<string, []>("input_197")];
            tensor<fp32, [1, 512, 3072]> input_199 = linear(bias = text_branch_encoder_layer_8_intermediate_dense_bias, weight = text_branch_encoder_layer_8_intermediate_dense_weight, x = input_197)[name = tensor<string, []>("linear_52")];
            tensor<string, []> input_201_mode_0 = const()[name = tensor<string, []>("input_201_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_201 = gelu(mode = input_201_mode_0, x = input_199)[name = tensor<string, []>("input_201")];
            tensor<fp32, [1, 512, 768]> input_203 = linear(bias = text_branch_encoder_layer_8_output_dense_bias, weight = text_branch_encoder_layer_8_output_dense_weight, x = input_201)[name = tensor<string, []>("linear_53")];
            tensor<fp32, [1, 512, 768]> input_205 = add(x = input_203, y = input_197)[name = tensor<string, []>("input_205")];
            tensor<int32, [1]> input_207_axes_0 = const()[name = tensor<string, []>("input_207_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_207 = layer_norm(axes = input_207_axes_0, beta = text_branch_encoder_layer_8_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_8_output_LayerNorm_weight, x = input_205)[name = tensor<string, []>("input_207")];
            tensor<fp32, [1, 512, 768]> x_117 = linear(bias = text_branch_encoder_layer_9_attention_self_query_bias, weight = text_branch_encoder_layer_9_attention_self_query_weight, x = input_207)[name = tensor<string, []>("linear_54")];
            tensor<fp32, [1, 512, 768]> x_109 = linear(bias = text_branch_encoder_layer_9_attention_self_key_bias, weight = text_branch_encoder_layer_9_attention_self_key_weight, x = input_207)[name = tensor<string, []>("linear_55")];
            tensor<int32, [4]> var_803 = const()[name = tensor<string, []>("op_803"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_111 = reshape(shape = var_803, x = x_109)[name = tensor<string, []>("x_111")];
            tensor<fp32, [1, 512, 768]> x_113 = linear(bias = text_branch_encoder_layer_9_attention_self_value_bias, weight = text_branch_encoder_layer_9_attention_self_value_weight, x = input_207)[name = tensor<string, []>("linear_56")];
            tensor<int32, [4]> var_812 = const()[name = tensor<string, []>("op_812"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_115 = reshape(shape = var_812, x = x_113)[name = tensor<string, []>("x_115")];
            tensor<int32, [4]> var_814 = const()[name = tensor<string, []>("op_814"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_818 = const()[name = tensor<string, []>("op_818"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_119 = reshape(shape = var_818, x = x_117)[name = tensor<string, []>("x_119")];
            tensor<bool, []> attention_scores_37_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_37_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = tensor<string, []>("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_55_perm_0 = const()[name = tensor<string, []>("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_55 = transpose(perm = transpose_55_perm_0, x = x_111)[name = tensor<string, []>("transpose_69")];
            tensor<fp32, [1, 12, 512, 64]> transpose_54 = transpose(perm = transpose_54_perm_0, x = x_119)[name = tensor<string, []>("transpose_70")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_37 = matmul(transpose_x = attention_scores_37_transpose_x_0, transpose_y = attention_scores_37_transpose_y_0, x = transpose_54, y = transpose_55)[name = tensor<string, []>("attention_scores_37")];
            tensor<fp32, []> _inversed_attention_scores_39_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_39_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_39 = mul(x = attention_scores_37, y = _inversed_attention_scores_39_y_0)[name = tensor<string, []>("_inversed_attention_scores_39")];
            tensor<fp32, [1, 12, 512, 512]> input_209 = add(x = _inversed_attention_scores_39, y = attention_mask_1)[name = tensor<string, []>("input_209")];
            tensor<fp32, [1, 12, 512, 512]> input_211 = softmax(axis = var_10, x = input_209)[name = tensor<string, []>("input_211")];
            tensor<bool, []> context_layer_37_transpose_x_0 = const()[name = tensor<string, []>("context_layer_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_37_transpose_y_0 = const()[name = tensor<string, []>("context_layer_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_19 = transpose(perm = var_814, x = x_115)[name = tensor<string, []>("transpose_71")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_37 = matmul(transpose_x = context_layer_37_transpose_x_0, transpose_y = context_layer_37_transpose_y_0, x = input_211, y = value_layer_19)[name = tensor<string, []>("context_layer_37")];
            tensor<int32, [4]> var_830 = const()[name = tensor<string, []>("op_830"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_835 = const()[name = tensor<string, []>("op_835"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_831 = transpose(perm = var_830, x = context_layer_37)[name = tensor<string, []>("transpose_68")];
            tensor<fp32, [1, 512, 768]> input_213 = reshape(shape = var_835, x = var_831)[name = tensor<string, []>("input_213")];
            tensor<fp32, [1, 512, 768]> input_215 = linear(bias = text_branch_encoder_layer_9_attention_output_dense_bias, weight = text_branch_encoder_layer_9_attention_output_dense_weight, x = input_213)[name = tensor<string, []>("linear_57")];
            tensor<fp32, [1, 512, 768]> input_217 = add(x = input_215, y = input_207)[name = tensor<string, []>("input_217")];
            tensor<int32, [1]> input_219_axes_0 = const()[name = tensor<string, []>("input_219_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_219 = layer_norm(axes = input_219_axes_0, beta = text_branch_encoder_layer_9_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_9_attention_output_LayerNorm_weight, x = input_217)[name = tensor<string, []>("input_219")];
            tensor<fp32, [1, 512, 3072]> input_221 = linear(bias = text_branch_encoder_layer_9_intermediate_dense_bias, weight = text_branch_encoder_layer_9_intermediate_dense_weight, x = input_219)[name = tensor<string, []>("linear_58")];
            tensor<string, []> input_223_mode_0 = const()[name = tensor<string, []>("input_223_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_223 = gelu(mode = input_223_mode_0, x = input_221)[name = tensor<string, []>("input_223")];
            tensor<fp32, [1, 512, 768]> input_225 = linear(bias = text_branch_encoder_layer_9_output_dense_bias, weight = text_branch_encoder_layer_9_output_dense_weight, x = input_223)[name = tensor<string, []>("linear_59")];
            tensor<fp32, [1, 512, 768]> input_227 = add(x = input_225, y = input_219)[name = tensor<string, []>("input_227")];
            tensor<int32, [1]> input_229_axes_0 = const()[name = tensor<string, []>("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_229 = layer_norm(axes = input_229_axes_0, beta = text_branch_encoder_layer_9_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_9_output_LayerNorm_weight, x = input_227)[name = tensor<string, []>("input_229")];
            tensor<fp32, [1, 512, 768]> x_129 = linear(bias = text_branch_encoder_layer_10_attention_self_query_bias, weight = text_branch_encoder_layer_10_attention_self_query_weight, x = input_229)[name = tensor<string, []>("linear_60")];
            tensor<fp32, [1, 512, 768]> x_121 = linear(bias = text_branch_encoder_layer_10_attention_self_key_bias, weight = text_branch_encoder_layer_10_attention_self_key_weight, x = input_229)[name = tensor<string, []>("linear_61")];
            tensor<int32, [4]> var_880 = const()[name = tensor<string, []>("op_880"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_123 = reshape(shape = var_880, x = x_121)[name = tensor<string, []>("x_123")];
            tensor<fp32, [1, 512, 768]> x_125 = linear(bias = text_branch_encoder_layer_10_attention_self_value_bias, weight = text_branch_encoder_layer_10_attention_self_value_weight, x = input_229)[name = tensor<string, []>("linear_62")];
            tensor<int32, [4]> var_889 = const()[name = tensor<string, []>("op_889"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_127 = reshape(shape = var_889, x = x_125)[name = tensor<string, []>("x_127")];
            tensor<int32, [4]> var_891 = const()[name = tensor<string, []>("op_891"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_895 = const()[name = tensor<string, []>("op_895"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_131 = reshape(shape = var_895, x = x_129)[name = tensor<string, []>("x_131")];
            tensor<bool, []> attention_scores_41_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_41_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = tensor<string, []>("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_57 = transpose(perm = transpose_57_perm_0, x = x_123)[name = tensor<string, []>("transpose_65")];
            tensor<fp32, [1, 12, 512, 64]> transpose_56 = transpose(perm = transpose_56_perm_0, x = x_131)[name = tensor<string, []>("transpose_66")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_41 = matmul(transpose_x = attention_scores_41_transpose_x_0, transpose_y = attention_scores_41_transpose_y_0, x = transpose_56, y = transpose_57)[name = tensor<string, []>("attention_scores_41")];
            tensor<fp32, []> _inversed_attention_scores_43_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_43_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores_43 = mul(x = attention_scores_41, y = _inversed_attention_scores_43_y_0)[name = tensor<string, []>("_inversed_attention_scores_43")];
            tensor<fp32, [1, 12, 512, 512]> input_231 = add(x = _inversed_attention_scores_43, y = attention_mask_1)[name = tensor<string, []>("input_231")];
            tensor<fp32, [1, 12, 512, 512]> input_233 = softmax(axis = var_10, x = input_231)[name = tensor<string, []>("input_233")];
            tensor<bool, []> context_layer_41_transpose_x_0 = const()[name = tensor<string, []>("context_layer_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_41_transpose_y_0 = const()[name = tensor<string, []>("context_layer_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer_21 = transpose(perm = var_891, x = x_127)[name = tensor<string, []>("transpose_67")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_41 = matmul(transpose_x = context_layer_41_transpose_x_0, transpose_y = context_layer_41_transpose_y_0, x = input_233, y = value_layer_21)[name = tensor<string, []>("context_layer_41")];
            tensor<int32, [4]> var_907 = const()[name = tensor<string, []>("op_907"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_912 = const()[name = tensor<string, []>("op_912"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_908 = transpose(perm = var_907, x = context_layer_41)[name = tensor<string, []>("transpose_64")];
            tensor<fp32, [1, 512, 768]> input_235 = reshape(shape = var_912, x = var_908)[name = tensor<string, []>("input_235")];
            tensor<fp32, [1, 512, 768]> input_237 = linear(bias = text_branch_encoder_layer_10_attention_output_dense_bias, weight = text_branch_encoder_layer_10_attention_output_dense_weight, x = input_235)[name = tensor<string, []>("linear_63")];
            tensor<fp32, [1, 512, 768]> input_239 = add(x = input_237, y = input_229)[name = tensor<string, []>("input_239")];
            tensor<int32, [1]> input_241_axes_0 = const()[name = tensor<string, []>("input_241_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_241 = layer_norm(axes = input_241_axes_0, beta = text_branch_encoder_layer_10_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_10_attention_output_LayerNorm_weight, x = input_239)[name = tensor<string, []>("input_241")];
            tensor<fp32, [1, 512, 3072]> input_243 = linear(bias = text_branch_encoder_layer_10_intermediate_dense_bias, weight = text_branch_encoder_layer_10_intermediate_dense_weight, x = input_241)[name = tensor<string, []>("linear_64")];
            tensor<string, []> input_245_mode_0 = const()[name = tensor<string, []>("input_245_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_245 = gelu(mode = input_245_mode_0, x = input_243)[name = tensor<string, []>("input_245")];
            tensor<fp32, [1, 512, 768]> input_247 = linear(bias = text_branch_encoder_layer_10_output_dense_bias, weight = text_branch_encoder_layer_10_output_dense_weight, x = input_245)[name = tensor<string, []>("linear_65")];
            tensor<fp32, [1, 512, 768]> input_249 = add(x = input_247, y = input_241)[name = tensor<string, []>("input_249")];
            tensor<int32, [1]> input_251_axes_0 = const()[name = tensor<string, []>("input_251_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_251 = layer_norm(axes = input_251_axes_0, beta = text_branch_encoder_layer_10_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_10_output_LayerNorm_weight, x = input_249)[name = tensor<string, []>("input_251")];
            tensor<fp32, [1, 512, 768]> x_141 = linear(bias = text_branch_encoder_layer_11_attention_self_query_bias, weight = text_branch_encoder_layer_11_attention_self_query_weight, x = input_251)[name = tensor<string, []>("linear_66")];
            tensor<fp32, [1, 512, 768]> x_133 = linear(bias = text_branch_encoder_layer_11_attention_self_key_bias, weight = text_branch_encoder_layer_11_attention_self_key_weight, x = input_251)[name = tensor<string, []>("linear_67")];
            tensor<int32, [4]> var_957 = const()[name = tensor<string, []>("op_957"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_135 = reshape(shape = var_957, x = x_133)[name = tensor<string, []>("x_135")];
            tensor<fp32, [1, 512, 768]> x_137 = linear(bias = text_branch_encoder_layer_11_attention_self_value_bias, weight = text_branch_encoder_layer_11_attention_self_value_weight, x = input_251)[name = tensor<string, []>("linear_68")];
            tensor<int32, [4]> var_966 = const()[name = tensor<string, []>("op_966"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x_139 = reshape(shape = var_966, x = x_137)[name = tensor<string, []>("x_139")];
            tensor<int32, [4]> var_968 = const()[name = tensor<string, []>("op_968"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_972 = const()[name = tensor<string, []>("op_972"), val = tensor<int32, [4]>([1, 512, 12, 64])];
            tensor<fp32, [1, 512, 12, 64]> x = reshape(shape = var_972, x = x_141)[name = tensor<string, []>("x")];
            tensor<bool, []> attention_scores_45_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_45_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp32, [1, 12, 64, 512]> transpose_59 = transpose(perm = transpose_59_perm_0, x = x_135)[name = tensor<string, []>("transpose_61")];
            tensor<fp32, [1, 12, 512, 64]> transpose_58 = transpose(perm = transpose_58_perm_0, x = x)[name = tensor<string, []>("transpose_62")];
            tensor<fp32, [1, 12, 512, 512]> attention_scores_45 = matmul(transpose_x = attention_scores_45_transpose_x_0, transpose_y = attention_scores_45_transpose_y_0, x = transpose_58, y = transpose_59)[name = tensor<string, []>("attention_scores_45")];
            tensor<fp32, []> _inversed_attention_scores_y_0 = const()[name = tensor<string, []>("_inversed_attention_scores_y_0"), val = tensor<fp32, []>(0x1p-3)];
            tensor<fp32, [1, 12, 512, 512]> _inversed_attention_scores = mul(x = attention_scores_45, y = _inversed_attention_scores_y_0)[name = tensor<string, []>("_inversed_attention_scores")];
            tensor<fp32, [1, 12, 512, 512]> input_253 = add(x = _inversed_attention_scores, y = attention_mask_1)[name = tensor<string, []>("input_253")];
            tensor<fp32, [1, 12, 512, 512]> input_255 = softmax(axis = var_10, x = input_253)[name = tensor<string, []>("input_255")];
            tensor<bool, []> context_layer_45_transpose_x_0 = const()[name = tensor<string, []>("context_layer_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_45_transpose_y_0 = const()[name = tensor<string, []>("context_layer_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp32, [1, 12, 512, 64]> value_layer = transpose(perm = var_968, x = x_139)[name = tensor<string, []>("transpose_63")];
            tensor<fp32, [1, 12, 512, 64]> context_layer_45 = matmul(transpose_x = context_layer_45_transpose_x_0, transpose_y = context_layer_45_transpose_y_0, x = input_255, y = value_layer)[name = tensor<string, []>("context_layer_45")];
            tensor<int32, [4]> var_984 = const()[name = tensor<string, []>("op_984"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_989 = const()[name = tensor<string, []>("op_989"), val = tensor<int32, [3]>([1, 512, 768])];
            tensor<fp32, [1, 512, 12, 64]> var_985 = transpose(perm = var_984, x = context_layer_45)[name = tensor<string, []>("transpose_60")];
            tensor<fp32, [1, 512, 768]> input_257 = reshape(shape = var_989, x = var_985)[name = tensor<string, []>("input_257")];
            tensor<fp32, [1, 512, 768]> input_259 = linear(bias = text_branch_encoder_layer_11_attention_output_dense_bias, weight = text_branch_encoder_layer_11_attention_output_dense_weight, x = input_257)[name = tensor<string, []>("linear_69")];
            tensor<fp32, [1, 512, 768]> input_261 = add(x = input_259, y = input_251)[name = tensor<string, []>("input_261")];
            tensor<int32, [1]> input_263_axes_0 = const()[name = tensor<string, []>("input_263_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> input_263 = layer_norm(axes = input_263_axes_0, beta = text_branch_encoder_layer_11_attention_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_11_attention_output_LayerNorm_weight, x = input_261)[name = tensor<string, []>("input_263")];
            tensor<fp32, [1, 512, 3072]> input_265 = linear(bias = text_branch_encoder_layer_11_intermediate_dense_bias, weight = text_branch_encoder_layer_11_intermediate_dense_weight, x = input_263)[name = tensor<string, []>("linear_70")];
            tensor<string, []> input_267_mode_0 = const()[name = tensor<string, []>("input_267_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp32, [1, 512, 3072]> input_267 = gelu(mode = input_267_mode_0, x = input_265)[name = tensor<string, []>("input_267")];
            tensor<fp32, [1, 512, 768]> input_269 = linear(bias = text_branch_encoder_layer_11_output_dense_bias, weight = text_branch_encoder_layer_11_output_dense_weight, x = input_267)[name = tensor<string, []>("linear_71")];
            tensor<fp32, [1, 512, 768]> input_271 = add(x = input_269, y = input_263)[name = tensor<string, []>("input_271")];
            tensor<int32, [1]> hidden_states_axes_0 = const()[name = tensor<string, []>("hidden_states_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp32, [1, 512, 768]> hidden_states = layer_norm(axes = hidden_states_axes_0, beta = text_branch_encoder_layer_11_output_LayerNorm_bias, epsilon = var_12, gamma = text_branch_encoder_layer_11_output_LayerNorm_weight, x = input_271)[name = tensor<string, []>("hidden_states")];
            tensor<int32, [3]> input_273_begin_0 = const()[name = tensor<string, []>("input_273_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> input_273_end_0 = const()[name = tensor<string, []>("input_273_end_0"), val = tensor<int32, [3]>([1, 1, 768])];
            tensor<bool, [3]> input_273_end_mask_0 = const()[name = tensor<string, []>("input_273_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> input_273_squeeze_mask_0 = const()[name = tensor<string, []>("input_273_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<fp32, [1, 768]> input_273 = slice_by_index(begin = input_273_begin_0, end = input_273_end_0, end_mask = input_273_end_mask_0, squeeze_mask = input_273_squeeze_mask_0, x = hidden_states)[name = tensor<string, []>("input_273")];
            tensor<fp32, [1, 768]> input_275 = linear(bias = text_branch_pooler_dense_bias, weight = text_branch_pooler_dense_weight, x = input_273)[name = tensor<string, []>("linear_72")];
            tensor<fp32, [1, 768]> input_277 = tanh(x = input_275)[name = tensor<string, []>("input_277")];
            tensor<fp32, [1, 512]> input_279 = linear(bias = text_projection_0_bias, weight = text_projection_0_weight, x = input_277)[name = tensor<string, []>("linear_73")];
            tensor<fp32, [1, 512]> input_281 = relu(x = input_279)[name = tensor<string, []>("input_281")];
            tensor<fp32, [1, 512]> input = linear(bias = text_projection_2_bias, weight = text_projection_2_weight, x = input_281)[name = tensor<string, []>("linear_74")];
            tensor<int32, [1]> var_1036 = const()[name = tensor<string, []>("op_1036"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1037 = const()[name = tensor<string, []>("op_1037"), val = tensor<bool, []>(true)];
            tensor<fp32, [1, 1]> var_1039 = reduce_l2_norm(axes = var_1036, keep_dims = var_1037, x = input)[name = tensor<string, []>("op_1039")];
            tensor<fp32, []> var_1040 = const()[name = tensor<string, []>("op_1040"), val = tensor<fp32, []>(0x1.197998p-40)];
            tensor<fp32, [1, 1]> var_1041 = maximum(x = var_1039, y = var_1040)[name = tensor<string, []>("op_1041")];
            tensor<int32, [2]> denom_reps_0 = const()[name = tensor<string, []>("denom_reps_0"), val = tensor<int32, [2]>([1, 512])];
            tensor<fp32, [1, 512]> denom = tile(reps = denom_reps_0, x = var_1041)[name = tensor<string, []>("denom")];
            tensor<fp32, [1, 512]> text_embedding = real_div(x = input, y = denom)[name = tensor<string, []>("op_1043")];
        } -> (text_embedding);
}