program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios17>(tensor<int32, [1, 512]> attention_mask, tensor<int32, [1, 512]> input_ids) {
            tensor<int32, []> var_14 = const()[name = tensor<string, []>("op_14"), val = tensor<int32, []>(1)];
            tensor<int32, []> var_21 = const()[name = tensor<string, []>("op_21"), val = tensor<int32, []>(-1)];
            tensor<bool, [1, 512]> var_34 = not_equal(x = input_ids, y = var_14)[name = tensor<string, []>("op_34")];
            tensor<string, []> mask_1_dtype_0 = const()[name = tensor<string, []>("mask_1_dtype_0"), val = tensor<string, []>("int32")];
            tensor<bool, []> var_36_exclusive_0 = const()[name = tensor<string, []>("op_36_exclusive_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_36_reverse_0 = const()[name = tensor<string, []>("op_36_reverse_0"), val = tensor<bool, []>(false)];
            tensor<int32, [1, 512]> mask_1 = cast(dtype = mask_1_dtype_0, x = var_34)[name = tensor<string, []>("cast_61")];
            tensor<int32, [1, 512]> var_36 = cumsum(axis = var_14, exclusive = var_36_exclusive_0, reverse = var_36_reverse_0, x = mask_1)[name = tensor<string, []>("op_36")];
            tensor<int32, [1, 512]> incremental_indices = mul(x = var_36, y = mask_1)[name = tensor<string, []>("incremental_indices")];
            tensor<int32, []> var_42 = const()[name = tensor<string, []>("op_42"), val = tensor<int32, []>(1)];
            tensor<int32, [1, 512]> position_ids = add(x = incremental_indices, y = var_42)[name = tensor<string, []>("position_ids")];
            tensor<bool, []> buffered_token_type_ids_validate_indices_0 = const()[name = tensor<string, []>("buffered_token_type_ids_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [1, 514]> const_2_to_uint16 = const()[name = tensor<string, []>("const_2_to_uint16"), val = tensor<uint16, [1, 514]>([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])];
            tensor<string, []> position_ids_to_uint16_dtype_0 = const()[name = tensor<string, []>("position_ids_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
            tensor<uint16, [1, 512]> position_ids_to_uint16 = cast(dtype = position_ids_to_uint16_dtype_0, x = position_ids)[name = tensor<string, []>("cast_60")];
            tensor<uint16, [1, 512]> buffered_token_type_ids_cast_uint16 = gather_along_axis(axis = var_14, indices = position_ids_to_uint16, validate_indices = buffered_token_type_ids_validate_indices_0, x = const_2_to_uint16)[name = tensor<string, []>("buffered_token_type_ids_cast_uint16")];
            tensor<int32, []> inputs_embeds_1_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_1_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> inputs_embeds_1_validate_indices_0 = const()[name = tensor<string, []>("inputs_embeds_1_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [50265, 768]> text_model_embeddings_word_embeddings_weight_to_fp16 = const()[name = tensor<string, []>("text_model_embeddings_word_embeddings_weight_to_fp16"), val = tensor<fp16, [50265, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            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_1_cast_fp16_axis_0 = const()[name = tensor<string, []>("inputs_embeds_1_cast_fp16_axis_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [1, 512, 768]> inputs_embeds_1_cast_fp16 = gather(axis = inputs_embeds_1_cast_fp16_axis_0, batch_dims = inputs_embeds_1_batch_dims_0, indices = select_0, validate_indices = inputs_embeds_1_validate_indices_0, x = text_model_embeddings_word_embeddings_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_1_cast_fp16")];
            tensor<int32, []> token_type_embeddings_1_axis_0 = const()[name = tensor<string, []>("token_type_embeddings_1_axis_0"), val = tensor<int32, []>(0)];
            tensor<int32, []> token_type_embeddings_1_batch_dims_0 = const()[name = tensor<string, []>("token_type_embeddings_1_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> token_type_embeddings_1_validate_indices_0 = const()[name = tensor<string, []>("token_type_embeddings_1_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 768]> text_model_embeddings_token_type_embeddings_weight_to_fp16 = const()[name = tensor<string, []>("text_model_embeddings_token_type_embeddings_weight_to_fp16"), val = tensor<fp16, [1, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77207168)))];
            tensor<fp16, [1, 512, 768]> token_type_embeddings_1_cast_fp16_cast_uint16 = gather(axis = token_type_embeddings_1_axis_0, batch_dims = token_type_embeddings_1_batch_dims_0, indices = buffered_token_type_ids_cast_uint16, validate_indices = token_type_embeddings_1_validate_indices_0, x = text_model_embeddings_token_type_embeddings_weight_to_fp16)[name = tensor<string, []>("token_type_embeddings_1_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 512, 768]> embeddings_1_cast_fp16 = add(x = inputs_embeds_1_cast_fp16, y = token_type_embeddings_1_cast_fp16_cast_uint16)[name = tensor<string, []>("embeddings_1_cast_fp16")];
            tensor<int32, []> position_embeddings_1_axis_0 = const()[name = tensor<string, []>("position_embeddings_1_axis_0"), val = tensor<int32, []>(0)];
            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<fp16, [514, 768]> text_model_embeddings_position_embeddings_weight_to_fp16 = const()[name = tensor<string, []>("text_model_embeddings_position_embeddings_weight_to_fp16"), val = tensor<fp16, [514, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77208768)))];
            tensor<fp16, [1, 512, 768]> position_embeddings_1_cast_fp16_cast_uint16 = gather(axis = position_embeddings_1_axis_0, batch_dims = position_embeddings_1_batch_dims_0, indices = position_ids_to_uint16, validate_indices = position_embeddings_1_validate_indices_0, x = text_model_embeddings_position_embeddings_weight_to_fp16)[name = tensor<string, []>("position_embeddings_1_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 512, 768]> input_3_cast_fp16 = add(x = embeddings_1_cast_fp16, y = position_embeddings_1_cast_fp16_cast_uint16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<int32, [1]> input_5_axes_0 = const()[name = tensor<string, []>("input_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_embeddings_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_embeddings_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77998336)))];
            tensor<fp16, [768]> text_model_embeddings_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_embeddings_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77999936)))];
            tensor<fp16, []> var_23_to_fp16 = const()[name = tensor<string, []>("op_23_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 512, 768]> input_5_cast_fp16 = layer_norm(axes = input_5_axes_0, beta = text_model_embeddings_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_embeddings_LayerNorm_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<string, []> attention_mask_3_dtype_0 = const()[name = tensor<string, []>("attention_mask_3_dtype_0"), val = tensor<string, []>("bool")];
            tensor<bool, [1, 1, 512, 1]> const_13 = const()[name = tensor<string, []>("const_13"), val = tensor<bool, [1, 1, 512, 1]>([[[[true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true], [true]]]])];
            tensor<string, []> cast_5_dtype_0 = const()[name = tensor<string, []>("cast_5_dtype_0"), val = tensor<string, []>("int8")];
            tensor<int32, []> gather_nd_0_batch_dims_0 = const()[name = tensor<string, []>("gather_nd_0_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_nd_0_validate_indices_0 = const()[name = tensor<string, []>("gather_nd_0_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [1, 1, 1, 512, 2]> stack_0_to_uint16 = const()[name = tensor<string, []>("stack_0_to_uint16"), val = tensor<uint16, [1, 1, 1, 512, 2]>([[[[[0, 0], [0, 1], [0, 2], [0, 3], [0, 4], [0, 5], [0, 6], [0, 7], [0, 8], [0, 9], [0, 10], [0, 11], [0, 12], [0, 13], [0, 14], [0, 15], [0, 16], [0, 17], [0, 18], [0, 19], [0, 20], [0, 21], [0, 22], [0, 23], [0, 24], [0, 25], [0, 26], [0, 27], [0, 28], [0, 29], [0, 30], [0, 31], [0, 32], [0, 33], [0, 34], [0, 35], [0, 36], [0, 37], [0, 38], [0, 39], [0, 40], [0, 41], [0, 42], [0, 43], [0, 44], [0, 45], [0, 46], [0, 47], [0, 48], [0, 49], [0, 50], [0, 51], [0, 52], [0, 53], [0, 54], [0, 55], [0, 56], [0, 57], [0, 58], [0, 59], [0, 60], [0, 61], [0, 62], [0, 63], [0, 64], [0, 65], [0, 66], [0, 67], [0, 68], [0, 69], [0, 70], [0, 71], [0, 72], [0, 73], [0, 74], [0, 75], [0, 76], [0, 77], [0, 78], [0, 79], [0, 80], [0, 81], [0, 82], [0, 83], [0, 84], [0, 85], [0, 86], [0, 87], [0, 88], [0, 89], [0, 90], [0, 91], [0, 92], [0, 93], [0, 94], [0, 95], [0, 96], [0, 97], [0, 98], [0, 99], [0, 100], [0, 101], [0, 102], [0, 103], [0, 104], [0, 105], [0, 106], [0, 107], [0, 108], [0, 109], [0, 110], [0, 111], [0, 112], [0, 113], [0, 114], [0, 115], [0, 116], [0, 117], [0, 118], [0, 119], [0, 120], [0, 121], [0, 122], [0, 123], [0, 124], [0, 125], [0, 126], [0, 127], [0, 128], [0, 129], [0, 130], [0, 131], [0, 132], [0, 133], [0, 134], [0, 135], [0, 136], [0, 137], [0, 138], [0, 139], [0, 140], [0, 141], [0, 142], [0, 143], [0, 144], [0, 145], [0, 146], [0, 147], [0, 148], [0, 149], [0, 150], [0, 151], [0, 152], [0, 153], [0, 154], [0, 155], [0, 156], [0, 157], [0, 158], [0, 159], [0, 160], [0, 161], [0, 162], [0, 163], [0, 164], [0, 165], [0, 166], [0, 167], [0, 168], [0, 169], [0, 170], [0, 171], [0, 172], [0, 173], [0, 174], [0, 175], [0, 176], [0, 177], [0, 178], [0, 179], [0, 180], [0, 181], [0, 182], [0, 183], [0, 184], [0, 185], [0, 186], [0, 187], [0, 188], [0, 189], [0, 190], [0, 191], [0, 192], [0, 193], [0, 194], [0, 195], [0, 196], [0, 197], [0, 198], [0, 199], [0, 200], [0, 201], [0, 202], [0, 203], [0, 204], [0, 205], [0, 206], [0, 207], [0, 208], [0, 209], [0, 210], [0, 211], [0, 212], [0, 213], [0, 214], [0, 215], [0, 216], [0, 217], [0, 218], [0, 219], [0, 220], [0, 221], [0, 222], [0, 223], [0, 224], [0, 225], [0, 226], [0, 227], [0, 228], [0, 229], [0, 230], [0, 231], [0, 232], [0, 233], [0, 234], [0, 235], [0, 236], [0, 237], [0, 238], [0, 239], [0, 240], [0, 241], [0, 242], [0, 243], [0, 244], [0, 245], [0, 246], [0, 247], [0, 248], [0, 249], [0, 250], [0, 251], [0, 252], [0, 253], [0, 254], [0, 255], [0, 256], [0, 257], [0, 258], [0, 259], [0, 260], [0, 261], [0, 262], [0, 263], [0, 264], [0, 265], [0, 266], [0, 267], [0, 268], [0, 269], [0, 270], [0, 271], [0, 272], [0, 273], [0, 274], [0, 275], [0, 276], [0, 277], [0, 278], [0, 279], [0, 280], [0, 281], [0, 282], [0, 283], [0, 284], [0, 285], [0, 286], [0, 287], [0, 288], [0, 289], [0, 290], [0, 291], [0, 292], [0, 293], [0, 294], [0, 295], [0, 296], [0, 297], [0, 298], [0, 299], [0, 300], [0, 301], [0, 302], [0, 303], [0, 304], [0, 305], [0, 306], [0, 307], [0, 308], [0, 309], [0, 310], [0, 311], [0, 312], [0, 313], [0, 314], [0, 315], [0, 316], [0, 317], [0, 318], [0, 319], [0, 320], [0, 321], [0, 322], [0, 323], [0, 324], [0, 325], [0, 326], [0, 327], [0, 328], [0, 329], [0, 330], [0, 331], [0, 332], [0, 333], [0, 334], [0, 335], [0, 336], [0, 337], [0, 338], [0, 339], [0, 340], [0, 341], [0, 342], [0, 343], [0, 344], [0, 345], [0, 346], [0, 347], [0, 348], [0, 349], [0, 350], [0, 351], [0, 352], [0, 353], [0, 354], [0, 355], [0, 356], [0, 357], [0, 358], [0, 359], [0, 360], [0, 361], [0, 362], [0, 363], [0, 364], [0, 365], [0, 366], [0, 367], [0, 368], [0, 369], [0, 370], [0, 371], [0, 372], [0, 373], [0, 374], [0, 375], [0, 376], [0, 377], [0, 378], [0, 379], [0, 380], [0, 381], [0, 382], [0, 383], [0, 384], [0, 385], [0, 386], [0, 387], [0, 388], [0, 389], [0, 390], [0, 391], [0, 392], [0, 393], [0, 394], [0, 395], [0, 396], [0, 397], [0, 398], [0, 399], [0, 400], [0, 401], [0, 402], [0, 403], [0, 404], [0, 405], [0, 406], [0, 407], [0, 408], [0, 409], [0, 410], [0, 411], [0, 412], [0, 413], [0, 414], [0, 415], [0, 416], [0, 417], [0, 418], [0, 419], [0, 420], [0, 421], [0, 422], [0, 423], [0, 424], [0, 425], [0, 426], [0, 427], [0, 428], [0, 429], [0, 430], [0, 431], [0, 432], [0, 433], [0, 434], [0, 435], [0, 436], [0, 437], [0, 438], [0, 439], [0, 440], [0, 441], [0, 442], [0, 443], [0, 444], [0, 445], [0, 446], [0, 447], [0, 448], [0, 449], [0, 450], [0, 451], [0, 452], [0, 453], [0, 454], [0, 455], [0, 456], [0, 457], [0, 458], [0, 459], [0, 460], [0, 461], [0, 462], [0, 463], [0, 464], [0, 465], [0, 466], [0, 467], [0, 468], [0, 469], [0, 470], [0, 471], [0, 472], [0, 473], [0, 474], [0, 475], [0, 476], [0, 477], [0, 478], [0, 479], [0, 480], [0, 481], [0, 482], [0, 483], [0, 484], [0, 485], [0, 486], [0, 487], [0, 488], [0, 489], [0, 490], [0, 491], [0, 492], [0, 493], [0, 494], [0, 495], [0, 496], [0, 497], [0, 498], [0, 499], [0, 500], [0, 501], [0, 502], [0, 503], [0, 504], [0, 505], [0, 506], [0, 507], [0, 508], [0, 509], [0, 510], [0, 511]]]]])];
            tensor<bool, [1, 512]> attention_mask_3 = cast(dtype = attention_mask_3_dtype_0, x = attention_mask)[name = tensor<string, []>("cast_59")];
            tensor<int8, [1, 512]> cast_5 = cast(dtype = cast_5_dtype_0, x = attention_mask_3)[name = tensor<string, []>("cast_58")];
            tensor<int8, [1, 1, 1, 512]> gather_nd_0_cast_uint16 = gather_nd(batch_dims = gather_nd_0_batch_dims_0, indices = stack_0_to_uint16, validate_indices = gather_nd_0_validate_indices_0, x = cast_5)[name = tensor<string, []>("gather_nd_0_cast_uint16")];
            tensor<string, []> var_95_transpose_dtype_0 = const()[name = tensor<string, []>("op_95_transpose_dtype_0"), val = tensor<string, []>("bool")];
            tensor<bool, [1, 1, 1, 512]> var_95_transpose = cast(dtype = var_95_transpose_dtype_0, x = gather_nd_0_cast_uint16)[name = tensor<string, []>("cast_57")];
            tensor<bool, [1, 1, 512, 512]> attention_mask_5 = logical_and(x = const_13, y = var_95_transpose)[name = tensor<string, []>("attention_mask_5")];
            tensor<fp16, [1, 1, 512, 512]> const_14_after_broadcast_to_fp16 = const()[name = tensor<string, []>("const_14_after_broadcast_to_fp16"), val = tensor<fp16, [1, 1, 512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78001536)))];
            tensor<fp16, [1, 1, 512, 512]> var_8_after_broadcast_to_fp16 = const()[name = tensor<string, []>("op_8_after_broadcast_to_fp16"), val = tensor<fp16, [1, 1, 512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78525888)))];
            tensor<fp16, [1, 1, 512, 512]> attention_mask_cast_fp16 = select(a = const_14_after_broadcast_to_fp16, b = var_8_after_broadcast_to_fp16, cond = attention_mask_5)[name = tensor<string, []>("attention_mask_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_0_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79050240)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80229952)))];
            tensor<fp16, [1, 512, 768]> linear_0_cast_fp16 = linear(bias = text_model_encoder_layer_0_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_0_attention_self_query_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<int32, [4]> var_140 = const()[name = tensor<string, []>("op_140"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_141_cast_fp16 = reshape(shape = var_140, x = linear_0_cast_fp16)[name = tensor<string, []>("op_141_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_0_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80231552)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81411264)))];
            tensor<fp16, [1, 512, 768]> linear_1_cast_fp16 = linear(bias = text_model_encoder_layer_0_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_0_attention_self_key_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<int32, [4]> var_146 = const()[name = tensor<string, []>("op_146"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_147_cast_fp16 = reshape(shape = var_146, x = linear_1_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_0_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81412864)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82592576)))];
            tensor<fp16, [1, 512, 768]> linear_2_cast_fp16 = linear(bias = text_model_encoder_layer_0_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_0_attention_self_value_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_152 = const()[name = tensor<string, []>("op_152"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_153_cast_fp16 = reshape(shape = var_152, x = linear_2_cast_fp16)[name = tensor<string, []>("op_153_cast_fp16")];
            tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_156_transpose_x_0 = const()[name = tensor<string, []>("op_156_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_156_transpose_y_0 = const()[name = tensor<string, []>("op_156_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_38 = transpose(perm = transpose_38_perm_0, x = var_147_cast_fp16)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [1, 12, 512, 64]> transpose_37 = transpose(perm = transpose_37_perm_0, x = var_141_cast_fp16)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 12, 512, 512]> var_156_cast_fp16 = matmul(transpose_x = var_156_transpose_x_0, transpose_y = var_156_transpose_y_0, x = transpose_37, y = transpose_38)[name = tensor<string, []>("op_156_cast_fp16")];
            tensor<fp16, []> var_157_to_fp16 = const()[name = tensor<string, []>("op_157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_1_cast_fp16 = mul(x = var_156_cast_fp16, y = var_157_to_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_7_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_160_cast_fp16 = softmax(axis = var_21, x = input_7_cast_fp16)[name = tensor<string, []>("op_160_cast_fp16")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = var_153_cast_fp16)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = var_160_cast_fp16, y = value_1_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_164_perm_0 = const()[name = tensor<string, []>("op_164_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_166 = const()[name = tensor<string, []>("op_166"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_164_cast_fp16 = transpose(perm = var_164_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 512, 768]> var_167_cast_fp16 = reshape(shape = var_166, x = var_164_cast_fp16)[name = tensor<string, []>("op_167_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_0_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82594176)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83773888)))];
            tensor<fp16, [1, 512, 768]> linear_3_cast_fp16 = linear(bias = text_model_encoder_layer_0_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_0_attention_output_dense_weight_to_fp16, x = var_167_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_15_cast_fp16 = add(x = linear_3_cast_fp16, y = input_5_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<int32, [1]> input_17_axes_0 = const()[name = tensor<string, []>("input_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83775488)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83777088)))];
            tensor<fp16, [1, 512, 768]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = text_model_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_0_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83778688)))];
            tensor<fp16, [3072]> text_model_encoder_layer_0_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88497344)))];
            tensor<fp16, [1, 512, 3072]> linear_4_cast_fp16 = linear(bias = text_model_encoder_layer_0_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_0_intermediate_dense_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> input_21_mode_0 = const()[name = tensor<string, []>("input_21_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_21_cast_fp16 = gelu(mode = input_21_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_0_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88503552)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93222208)))];
            tensor<fp16, [1, 512, 768]> linear_5_cast_fp16 = linear(bias = text_model_encoder_layer_0_output_dense_bias_to_fp16, weight = text_model_encoder_layer_0_output_dense_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_25_cast_fp16 = add(x = linear_5_cast_fp16, y = input_17_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<int32, [1]> hidden_states_5_axes_0 = const()[name = tensor<string, []>("hidden_states_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_0_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93223808)))];
            tensor<fp16, [768]> text_model_encoder_layer_0_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_0_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93225408)))];
            tensor<fp16, [1, 512, 768]> hidden_states_5_cast_fp16 = layer_norm(axes = hidden_states_5_axes_0, beta = text_model_encoder_layer_0_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_0_output_LayerNorm_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_1_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93227008)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94406720)))];
            tensor<fp16, [1, 512, 768]> linear_6_cast_fp16 = linear(bias = text_model_encoder_layer_1_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_1_attention_self_query_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<int32, [4]> var_209 = const()[name = tensor<string, []>("op_209"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_210_cast_fp16 = reshape(shape = var_209, x = linear_6_cast_fp16)[name = tensor<string, []>("op_210_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_1_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94408320)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95588032)))];
            tensor<fp16, [1, 512, 768]> linear_7_cast_fp16 = linear(bias = text_model_encoder_layer_1_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_1_attention_self_key_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<int32, [4]> var_215 = const()[name = tensor<string, []>("op_215"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_216_cast_fp16 = reshape(shape = var_215, x = linear_7_cast_fp16)[name = tensor<string, []>("op_216_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_1_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95589632)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96769344)))];
            tensor<fp16, [1, 512, 768]> linear_8_cast_fp16 = linear(bias = text_model_encoder_layer_1_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_1_attention_self_value_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<int32, [4]> var_221 = const()[name = tensor<string, []>("op_221"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_222_cast_fp16 = reshape(shape = var_221, x = linear_8_cast_fp16)[name = tensor<string, []>("op_222_cast_fp16")];
            tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_225_transpose_x_0 = const()[name = tensor<string, []>("op_225_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_225_transpose_y_0 = const()[name = tensor<string, []>("op_225_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_39_perm_0 = const()[name = tensor<string, []>("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_40_perm_0 = const()[name = tensor<string, []>("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_40 = transpose(perm = transpose_40_perm_0, x = var_216_cast_fp16)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 12, 512, 64]> transpose_39 = transpose(perm = transpose_39_perm_0, x = var_210_cast_fp16)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 12, 512, 512]> var_225_cast_fp16 = matmul(transpose_x = var_225_transpose_x_0, transpose_y = var_225_transpose_y_0, x = transpose_39, y = transpose_40)[name = tensor<string, []>("op_225_cast_fp16")];
            tensor<fp16, []> var_226_to_fp16 = const()[name = tensor<string, []>("op_226_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_5_cast_fp16 = mul(x = var_225_cast_fp16, y = var_226_to_fp16)[name = tensor<string, []>("attn_weights_5_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_27_cast_fp16 = add(x = attn_weights_5_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_229_cast_fp16 = softmax(axis = var_21, x = input_27_cast_fp16)[name = tensor<string, []>("op_229_cast_fp16")];
            tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = var_222_cast_fp16)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = var_229_cast_fp16, y = value_3_cast_fp16)[name = tensor<string, []>("attn_output_5_cast_fp16")];
            tensor<int32, [4]> var_233_perm_0 = const()[name = tensor<string, []>("op_233_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_235 = const()[name = tensor<string, []>("op_235"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_233_cast_fp16 = transpose(perm = var_233_perm_0, x = attn_output_5_cast_fp16)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 512, 768]> var_236_cast_fp16 = reshape(shape = var_235, x = var_233_cast_fp16)[name = tensor<string, []>("op_236_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_1_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96770944)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97950656)))];
            tensor<fp16, [1, 512, 768]> linear_9_cast_fp16 = linear(bias = text_model_encoder_layer_1_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_1_attention_output_dense_weight_to_fp16, x = var_236_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_35_cast_fp16 = add(x = linear_9_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<int32, [1]> input_37_axes_0 = const()[name = tensor<string, []>("input_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97952256)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97953856)))];
            tensor<fp16, [1, 512, 768]> input_37_cast_fp16 = layer_norm(axes = input_37_axes_0, beta = text_model_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_1_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97955456)))];
            tensor<fp16, [3072]> text_model_encoder_layer_1_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102674112)))];
            tensor<fp16, [1, 512, 3072]> linear_10_cast_fp16 = linear(bias = text_model_encoder_layer_1_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_1_intermediate_dense_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<string, []> input_41_mode_0 = const()[name = tensor<string, []>("input_41_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_41_cast_fp16 = gelu(mode = input_41_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_1_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102680320)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107398976)))];
            tensor<fp16, [1, 512, 768]> linear_11_cast_fp16 = linear(bias = text_model_encoder_layer_1_output_dense_bias_to_fp16, weight = text_model_encoder_layer_1_output_dense_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_45_cast_fp16 = add(x = linear_11_cast_fp16, y = input_37_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<int32, [1]> hidden_states_11_axes_0 = const()[name = tensor<string, []>("hidden_states_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_1_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107400576)))];
            tensor<fp16, [768]> text_model_encoder_layer_1_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_1_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107402176)))];
            tensor<fp16, [1, 512, 768]> hidden_states_11_cast_fp16 = layer_norm(axes = hidden_states_11_axes_0, beta = text_model_encoder_layer_1_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_1_output_LayerNorm_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_2_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107403776)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108583488)))];
            tensor<fp16, [1, 512, 768]> linear_12_cast_fp16 = linear(bias = text_model_encoder_layer_2_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_2_attention_self_query_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<int32, [4]> var_278 = const()[name = tensor<string, []>("op_278"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_279_cast_fp16 = reshape(shape = var_278, x = linear_12_cast_fp16)[name = tensor<string, []>("op_279_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_2_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108585088)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109764800)))];
            tensor<fp16, [1, 512, 768]> linear_13_cast_fp16 = linear(bias = text_model_encoder_layer_2_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_2_attention_self_key_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<int32, [4]> var_284 = const()[name = tensor<string, []>("op_284"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_285_cast_fp16 = reshape(shape = var_284, x = linear_13_cast_fp16)[name = tensor<string, []>("op_285_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_2_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109766400)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110946112)))];
            tensor<fp16, [1, 512, 768]> linear_14_cast_fp16 = linear(bias = text_model_encoder_layer_2_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_2_attention_self_value_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_290 = const()[name = tensor<string, []>("op_290"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_291_cast_fp16 = reshape(shape = var_290, x = linear_14_cast_fp16)[name = tensor<string, []>("op_291_cast_fp16")];
            tensor<int32, [4]> value_5_perm_0 = const()[name = tensor<string, []>("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_294_transpose_x_0 = const()[name = tensor<string, []>("op_294_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_294_transpose_y_0 = const()[name = tensor<string, []>("op_294_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_41_perm_0 = const()[name = tensor<string, []>("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_42_perm_0 = const()[name = tensor<string, []>("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_42 = transpose(perm = transpose_42_perm_0, x = var_285_cast_fp16)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 12, 512, 64]> transpose_41 = transpose(perm = transpose_41_perm_0, x = var_279_cast_fp16)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 12, 512, 512]> var_294_cast_fp16 = matmul(transpose_x = var_294_transpose_x_0, transpose_y = var_294_transpose_y_0, x = transpose_41, y = transpose_42)[name = tensor<string, []>("op_294_cast_fp16")];
            tensor<fp16, []> var_295_to_fp16 = const()[name = tensor<string, []>("op_295_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_9_cast_fp16 = mul(x = var_294_cast_fp16, y = var_295_to_fp16)[name = tensor<string, []>("attn_weights_9_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_47_cast_fp16 = add(x = attn_weights_9_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_298_cast_fp16 = softmax(axis = var_21, x = input_47_cast_fp16)[name = tensor<string, []>("op_298_cast_fp16")];
            tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = var_291_cast_fp16)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = var_298_cast_fp16, y = value_5_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")];
            tensor<int32, [4]> var_302_perm_0 = const()[name = tensor<string, []>("op_302_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_304 = const()[name = tensor<string, []>("op_304"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_302_cast_fp16 = transpose(perm = var_302_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 512, 768]> var_305_cast_fp16 = reshape(shape = var_304, x = var_302_cast_fp16)[name = tensor<string, []>("op_305_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_2_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110947712)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112127424)))];
            tensor<fp16, [1, 512, 768]> linear_15_cast_fp16 = linear(bias = text_model_encoder_layer_2_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_2_attention_output_dense_weight_to_fp16, x = var_305_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_55_cast_fp16 = add(x = linear_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<int32, [1]> input_57_axes_0 = const()[name = tensor<string, []>("input_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112129024)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112130624)))];
            tensor<fp16, [1, 512, 768]> input_57_cast_fp16 = layer_norm(axes = input_57_axes_0, beta = text_model_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_2_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112132224)))];
            tensor<fp16, [3072]> text_model_encoder_layer_2_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116850880)))];
            tensor<fp16, [1, 512, 3072]> linear_16_cast_fp16 = linear(bias = text_model_encoder_layer_2_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_2_intermediate_dense_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<string, []> input_61_mode_0 = const()[name = tensor<string, []>("input_61_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_2_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116857088)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121575744)))];
            tensor<fp16, [1, 512, 768]> linear_17_cast_fp16 = linear(bias = text_model_encoder_layer_2_output_dense_bias_to_fp16, weight = text_model_encoder_layer_2_output_dense_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_65_cast_fp16 = add(x = linear_17_cast_fp16, y = input_57_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<int32, [1]> hidden_states_17_axes_0 = const()[name = tensor<string, []>("hidden_states_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_2_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121577344)))];
            tensor<fp16, [768]> text_model_encoder_layer_2_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_2_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121578944)))];
            tensor<fp16, [1, 512, 768]> hidden_states_17_cast_fp16 = layer_norm(axes = hidden_states_17_axes_0, beta = text_model_encoder_layer_2_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_2_output_LayerNorm_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_3_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121580544)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122760256)))];
            tensor<fp16, [1, 512, 768]> linear_18_cast_fp16 = linear(bias = text_model_encoder_layer_3_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_3_attention_self_query_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<int32, [4]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_348_cast_fp16 = reshape(shape = var_347, x = linear_18_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_3_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122761856)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123941568)))];
            tensor<fp16, [1, 512, 768]> linear_19_cast_fp16 = linear(bias = text_model_encoder_layer_3_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_3_attention_self_key_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<int32, [4]> var_353 = const()[name = tensor<string, []>("op_353"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_354_cast_fp16 = reshape(shape = var_353, x = linear_19_cast_fp16)[name = tensor<string, []>("op_354_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_3_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123943168)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125122880)))];
            tensor<fp16, [1, 512, 768]> linear_20_cast_fp16 = linear(bias = text_model_encoder_layer_3_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_3_attention_self_value_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_359 = const()[name = tensor<string, []>("op_359"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_360_cast_fp16 = reshape(shape = var_359, x = linear_20_cast_fp16)[name = tensor<string, []>("op_360_cast_fp16")];
            tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_363_transpose_x_0 = const()[name = tensor<string, []>("op_363_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_363_transpose_y_0 = const()[name = tensor<string, []>("op_363_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_43_perm_0 = const()[name = tensor<string, []>("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_44_perm_0 = const()[name = tensor<string, []>("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_44 = transpose(perm = transpose_44_perm_0, x = var_354_cast_fp16)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [1, 12, 512, 64]> transpose_43 = transpose(perm = transpose_43_perm_0, x = var_348_cast_fp16)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [1, 12, 512, 512]> var_363_cast_fp16 = matmul(transpose_x = var_363_transpose_x_0, transpose_y = var_363_transpose_y_0, x = transpose_43, y = transpose_44)[name = tensor<string, []>("op_363_cast_fp16")];
            tensor<fp16, []> var_364_to_fp16 = const()[name = tensor<string, []>("op_364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_13_cast_fp16 = mul(x = var_363_cast_fp16, y = var_364_to_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_67_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_367_cast_fp16 = softmax(axis = var_21, x = input_67_cast_fp16)[name = tensor<string, []>("op_367_cast_fp16")];
            tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = var_360_cast_fp16)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = var_367_cast_fp16, y = value_7_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")];
            tensor<int32, [4]> var_371_perm_0 = const()[name = tensor<string, []>("op_371_perm_0"), 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, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_371_cast_fp16 = transpose(perm = var_371_perm_0, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [1, 512, 768]> var_374_cast_fp16 = reshape(shape = var_373, x = var_371_cast_fp16)[name = tensor<string, []>("op_374_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_3_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125124480)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126304192)))];
            tensor<fp16, [1, 512, 768]> linear_21_cast_fp16 = linear(bias = text_model_encoder_layer_3_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_3_attention_output_dense_weight_to_fp16, x = var_374_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_75_cast_fp16 = add(x = linear_21_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126305792)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126307392)))];
            tensor<fp16, [1, 512, 768]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_model_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_3_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126308992)))];
            tensor<fp16, [3072]> text_model_encoder_layer_3_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131027648)))];
            tensor<fp16, [1, 512, 3072]> linear_22_cast_fp16 = linear(bias = text_model_encoder_layer_3_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_3_intermediate_dense_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<string, []> input_81_mode_0 = const()[name = tensor<string, []>("input_81_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_81_cast_fp16 = gelu(mode = input_81_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_3_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131033856)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135752512)))];
            tensor<fp16, [1, 512, 768]> linear_23_cast_fp16 = linear(bias = text_model_encoder_layer_3_output_dense_bias_to_fp16, weight = text_model_encoder_layer_3_output_dense_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_85_cast_fp16 = add(x = linear_23_cast_fp16, y = input_77_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<int32, [1]> hidden_states_23_axes_0 = const()[name = tensor<string, []>("hidden_states_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_3_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135754112)))];
            tensor<fp16, [768]> text_model_encoder_layer_3_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_3_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135755712)))];
            tensor<fp16, [1, 512, 768]> hidden_states_23_cast_fp16 = layer_norm(axes = hidden_states_23_axes_0, beta = text_model_encoder_layer_3_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_3_output_LayerNorm_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_4_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135757312)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136937024)))];
            tensor<fp16, [1, 512, 768]> linear_24_cast_fp16 = linear(bias = text_model_encoder_layer_4_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_4_attention_self_query_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<int32, [4]> var_416 = const()[name = tensor<string, []>("op_416"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_417_cast_fp16 = reshape(shape = var_416, x = linear_24_cast_fp16)[name = tensor<string, []>("op_417_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_4_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136938624)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138118336)))];
            tensor<fp16, [1, 512, 768]> linear_25_cast_fp16 = linear(bias = text_model_encoder_layer_4_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_4_attention_self_key_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<int32, [4]> var_422 = const()[name = tensor<string, []>("op_422"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_423_cast_fp16 = reshape(shape = var_422, x = linear_25_cast_fp16)[name = tensor<string, []>("op_423_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_4_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138119936)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139299648)))];
            tensor<fp16, [1, 512, 768]> linear_26_cast_fp16 = linear(bias = text_model_encoder_layer_4_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_4_attention_self_value_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_428 = const()[name = tensor<string, []>("op_428"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_429_cast_fp16 = reshape(shape = var_428, x = linear_26_cast_fp16)[name = tensor<string, []>("op_429_cast_fp16")];
            tensor<int32, [4]> value_9_perm_0 = const()[name = tensor<string, []>("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_432_transpose_x_0 = const()[name = tensor<string, []>("op_432_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_432_transpose_y_0 = const()[name = tensor<string, []>("op_432_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_45_perm_0 = const()[name = tensor<string, []>("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_46_perm_0 = const()[name = tensor<string, []>("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_46 = transpose(perm = transpose_46_perm_0, x = var_423_cast_fp16)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [1, 12, 512, 64]> transpose_45 = transpose(perm = transpose_45_perm_0, x = var_417_cast_fp16)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [1, 12, 512, 512]> var_432_cast_fp16 = matmul(transpose_x = var_432_transpose_x_0, transpose_y = var_432_transpose_y_0, x = transpose_45, y = transpose_46)[name = tensor<string, []>("op_432_cast_fp16")];
            tensor<fp16, []> var_433_to_fp16 = const()[name = tensor<string, []>("op_433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_17_cast_fp16 = mul(x = var_432_cast_fp16, y = var_433_to_fp16)[name = tensor<string, []>("attn_weights_17_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_87_cast_fp16 = add(x = attn_weights_17_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_436_cast_fp16 = softmax(axis = var_21, x = input_87_cast_fp16)[name = tensor<string, []>("op_436_cast_fp16")];
            tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = var_429_cast_fp16)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = var_436_cast_fp16, y = value_9_cast_fp16)[name = tensor<string, []>("attn_output_17_cast_fp16")];
            tensor<int32, [4]> var_440_perm_0 = const()[name = tensor<string, []>("op_440_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_442 = const()[name = tensor<string, []>("op_442"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_440_cast_fp16 = transpose(perm = var_440_perm_0, x = attn_output_17_cast_fp16)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [1, 512, 768]> var_443_cast_fp16 = reshape(shape = var_442, x = var_440_cast_fp16)[name = tensor<string, []>("op_443_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_4_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139301248)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140480960)))];
            tensor<fp16, [1, 512, 768]> linear_27_cast_fp16 = linear(bias = text_model_encoder_layer_4_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_4_attention_output_dense_weight_to_fp16, x = var_443_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_95_cast_fp16 = add(x = linear_27_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<int32, [1]> input_97_axes_0 = const()[name = tensor<string, []>("input_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140482560)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140484160)))];
            tensor<fp16, [1, 512, 768]> input_97_cast_fp16 = layer_norm(axes = input_97_axes_0, beta = text_model_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_4_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140485760)))];
            tensor<fp16, [3072]> text_model_encoder_layer_4_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145204416)))];
            tensor<fp16, [1, 512, 3072]> linear_28_cast_fp16 = linear(bias = text_model_encoder_layer_4_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_4_intermediate_dense_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<string, []> input_101_mode_0 = const()[name = tensor<string, []>("input_101_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_101_cast_fp16 = gelu(mode = input_101_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_4_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145210624)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149929280)))];
            tensor<fp16, [1, 512, 768]> linear_29_cast_fp16 = linear(bias = text_model_encoder_layer_4_output_dense_bias_to_fp16, weight = text_model_encoder_layer_4_output_dense_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_105_cast_fp16 = add(x = linear_29_cast_fp16, y = input_97_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<int32, [1]> hidden_states_29_axes_0 = const()[name = tensor<string, []>("hidden_states_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_4_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149930880)))];
            tensor<fp16, [768]> text_model_encoder_layer_4_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_4_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149932480)))];
            tensor<fp16, [1, 512, 768]> hidden_states_29_cast_fp16 = layer_norm(axes = hidden_states_29_axes_0, beta = text_model_encoder_layer_4_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_4_output_LayerNorm_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("hidden_states_29_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_5_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149934080)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151113792)))];
            tensor<fp16, [1, 512, 768]> linear_30_cast_fp16 = linear(bias = text_model_encoder_layer_5_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_5_attention_self_query_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<int32, [4]> var_485 = const()[name = tensor<string, []>("op_485"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_486_cast_fp16 = reshape(shape = var_485, x = linear_30_cast_fp16)[name = tensor<string, []>("op_486_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_5_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151115392)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152295104)))];
            tensor<fp16, [1, 512, 768]> linear_31_cast_fp16 = linear(bias = text_model_encoder_layer_5_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_5_attention_self_key_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<int32, [4]> var_491 = const()[name = tensor<string, []>("op_491"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_492_cast_fp16 = reshape(shape = var_491, x = linear_31_cast_fp16)[name = tensor<string, []>("op_492_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_5_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152296704)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153476416)))];
            tensor<fp16, [1, 512, 768]> linear_32_cast_fp16 = linear(bias = text_model_encoder_layer_5_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_5_attention_self_value_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_497 = const()[name = tensor<string, []>("op_497"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_498_cast_fp16 = reshape(shape = var_497, x = linear_32_cast_fp16)[name = tensor<string, []>("op_498_cast_fp16")];
            tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_501_transpose_x_0 = const()[name = tensor<string, []>("op_501_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_501_transpose_y_0 = const()[name = tensor<string, []>("op_501_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_47_perm_0 = const()[name = tensor<string, []>("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = tensor<string, []>("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_48 = transpose(perm = transpose_48_perm_0, x = var_492_cast_fp16)[name = tensor<string, []>("transpose_86")];
            tensor<fp16, [1, 12, 512, 64]> transpose_47 = transpose(perm = transpose_47_perm_0, x = var_486_cast_fp16)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [1, 12, 512, 512]> var_501_cast_fp16 = matmul(transpose_x = var_501_transpose_x_0, transpose_y = var_501_transpose_y_0, x = transpose_47, y = transpose_48)[name = tensor<string, []>("op_501_cast_fp16")];
            tensor<fp16, []> var_502_to_fp16 = const()[name = tensor<string, []>("op_502_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_21_cast_fp16 = mul(x = var_501_cast_fp16, y = var_502_to_fp16)[name = tensor<string, []>("attn_weights_21_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_107_cast_fp16 = add(x = attn_weights_21_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_505_cast_fp16 = softmax(axis = var_21, x = input_107_cast_fp16)[name = tensor<string, []>("op_505_cast_fp16")];
            tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = var_498_cast_fp16)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = var_505_cast_fp16, y = value_11_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")];
            tensor<int32, [4]> var_509_perm_0 = const()[name = tensor<string, []>("op_509_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_509_cast_fp16 = transpose(perm = var_509_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_85")];
            tensor<fp16, [1, 512, 768]> var_512_cast_fp16 = reshape(shape = var_511, x = var_509_cast_fp16)[name = tensor<string, []>("op_512_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_5_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153478016)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154657728)))];
            tensor<fp16, [1, 512, 768]> linear_33_cast_fp16 = linear(bias = text_model_encoder_layer_5_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_5_attention_output_dense_weight_to_fp16, x = var_512_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_115_cast_fp16 = add(x = linear_33_cast_fp16, y = hidden_states_29_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<int32, [1]> input_117_axes_0 = const()[name = tensor<string, []>("input_117_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154659328)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154660928)))];
            tensor<fp16, [1, 512, 768]> input_117_cast_fp16 = layer_norm(axes = input_117_axes_0, beta = text_model_encoder_layer_5_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_5_attention_output_LayerNorm_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_5_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154662528)))];
            tensor<fp16, [3072]> text_model_encoder_layer_5_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159381184)))];
            tensor<fp16, [1, 512, 3072]> linear_34_cast_fp16 = linear(bias = text_model_encoder_layer_5_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_5_intermediate_dense_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<string, []> input_121_mode_0 = const()[name = tensor<string, []>("input_121_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_121_cast_fp16 = gelu(mode = input_121_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_5_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159387392)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164106048)))];
            tensor<fp16, [1, 512, 768]> linear_35_cast_fp16 = linear(bias = text_model_encoder_layer_5_output_dense_bias_to_fp16, weight = text_model_encoder_layer_5_output_dense_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_125_cast_fp16 = add(x = linear_35_cast_fp16, y = input_117_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<int32, [1]> hidden_states_35_axes_0 = const()[name = tensor<string, []>("hidden_states_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_5_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164107648)))];
            tensor<fp16, [768]> text_model_encoder_layer_5_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_5_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164109248)))];
            tensor<fp16, [1, 512, 768]> hidden_states_35_cast_fp16 = layer_norm(axes = hidden_states_35_axes_0, beta = text_model_encoder_layer_5_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_5_output_LayerNorm_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("hidden_states_35_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_6_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164110848)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165290560)))];
            tensor<fp16, [1, 512, 768]> linear_36_cast_fp16 = linear(bias = text_model_encoder_layer_6_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_6_attention_self_query_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<int32, [4]> var_554 = const()[name = tensor<string, []>("op_554"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_555_cast_fp16 = reshape(shape = var_554, x = linear_36_cast_fp16)[name = tensor<string, []>("op_555_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_6_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165292160)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166471872)))];
            tensor<fp16, [1, 512, 768]> linear_37_cast_fp16 = linear(bias = text_model_encoder_layer_6_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_6_attention_self_key_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<int32, [4]> var_560 = const()[name = tensor<string, []>("op_560"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_561_cast_fp16 = reshape(shape = var_560, x = linear_37_cast_fp16)[name = tensor<string, []>("op_561_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_6_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166473472)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167653184)))];
            tensor<fp16, [1, 512, 768]> linear_38_cast_fp16 = linear(bias = text_model_encoder_layer_6_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_6_attention_self_value_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_566 = const()[name = tensor<string, []>("op_566"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_567_cast_fp16 = reshape(shape = var_566, x = linear_38_cast_fp16)[name = tensor<string, []>("op_567_cast_fp16")];
            tensor<int32, [4]> value_13_perm_0 = const()[name = tensor<string, []>("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_570_transpose_x_0 = const()[name = tensor<string, []>("op_570_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_570_transpose_y_0 = const()[name = tensor<string, []>("op_570_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_49_perm_0 = const()[name = tensor<string, []>("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_50_perm_0 = const()[name = tensor<string, []>("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_50 = transpose(perm = transpose_50_perm_0, x = var_561_cast_fp16)[name = tensor<string, []>("transpose_82")];
            tensor<fp16, [1, 12, 512, 64]> transpose_49 = transpose(perm = transpose_49_perm_0, x = var_555_cast_fp16)[name = tensor<string, []>("transpose_83")];
            tensor<fp16, [1, 12, 512, 512]> var_570_cast_fp16 = matmul(transpose_x = var_570_transpose_x_0, transpose_y = var_570_transpose_y_0, x = transpose_49, y = transpose_50)[name = tensor<string, []>("op_570_cast_fp16")];
            tensor<fp16, []> var_571_to_fp16 = const()[name = tensor<string, []>("op_571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_25_cast_fp16 = mul(x = var_570_cast_fp16, y = var_571_to_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_127_cast_fp16 = add(x = attn_weights_25_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_574_cast_fp16 = softmax(axis = var_21, x = input_127_cast_fp16)[name = tensor<string, []>("op_574_cast_fp16")];
            tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = var_567_cast_fp16)[name = tensor<string, []>("transpose_84")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = var_574_cast_fp16, y = value_13_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")];
            tensor<int32, [4]> var_578_perm_0 = const()[name = tensor<string, []>("op_578_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_580 = const()[name = tensor<string, []>("op_580"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_578_cast_fp16 = transpose(perm = var_578_perm_0, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_81")];
            tensor<fp16, [1, 512, 768]> var_581_cast_fp16 = reshape(shape = var_580, x = var_578_cast_fp16)[name = tensor<string, []>("op_581_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_6_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167654784)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168834496)))];
            tensor<fp16, [1, 512, 768]> linear_39_cast_fp16 = linear(bias = text_model_encoder_layer_6_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_6_attention_output_dense_weight_to_fp16, x = var_581_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_135_cast_fp16 = add(x = linear_39_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<int32, [1]> input_137_axes_0 = const()[name = tensor<string, []>("input_137_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168836096)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168837696)))];
            tensor<fp16, [1, 512, 768]> input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = text_model_encoder_layer_6_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_6_attention_output_LayerNorm_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_6_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168839296)))];
            tensor<fp16, [3072]> text_model_encoder_layer_6_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173557952)))];
            tensor<fp16, [1, 512, 3072]> linear_40_cast_fp16 = linear(bias = text_model_encoder_layer_6_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_6_intermediate_dense_weight_to_fp16, x = input_137_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<string, []> input_141_mode_0 = const()[name = tensor<string, []>("input_141_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_141_cast_fp16 = gelu(mode = input_141_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_6_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173564160)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178282816)))];
            tensor<fp16, [1, 512, 768]> linear_41_cast_fp16 = linear(bias = text_model_encoder_layer_6_output_dense_bias_to_fp16, weight = text_model_encoder_layer_6_output_dense_weight_to_fp16, x = input_141_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_145_cast_fp16 = add(x = linear_41_cast_fp16, y = input_137_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<int32, [1]> hidden_states_41_axes_0 = const()[name = tensor<string, []>("hidden_states_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_6_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178284416)))];
            tensor<fp16, [768]> text_model_encoder_layer_6_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_6_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178286016)))];
            tensor<fp16, [1, 512, 768]> hidden_states_41_cast_fp16 = layer_norm(axes = hidden_states_41_axes_0, beta = text_model_encoder_layer_6_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_6_output_LayerNorm_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("hidden_states_41_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_7_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178287616)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179467328)))];
            tensor<fp16, [1, 512, 768]> linear_42_cast_fp16 = linear(bias = text_model_encoder_layer_7_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_7_attention_self_query_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<int32, [4]> var_623 = const()[name = tensor<string, []>("op_623"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_624_cast_fp16 = reshape(shape = var_623, x = linear_42_cast_fp16)[name = tensor<string, []>("op_624_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_7_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179468928)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180648640)))];
            tensor<fp16, [1, 512, 768]> linear_43_cast_fp16 = linear(bias = text_model_encoder_layer_7_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_7_attention_self_key_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<int32, [4]> var_629 = const()[name = tensor<string, []>("op_629"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_630_cast_fp16 = reshape(shape = var_629, x = linear_43_cast_fp16)[name = tensor<string, []>("op_630_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_7_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180650240)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181829952)))];
            tensor<fp16, [1, 512, 768]> linear_44_cast_fp16 = linear(bias = text_model_encoder_layer_7_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_7_attention_self_value_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_636_cast_fp16 = reshape(shape = var_635, x = linear_44_cast_fp16)[name = tensor<string, []>("op_636_cast_fp16")];
            tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_639_transpose_x_0 = const()[name = tensor<string, []>("op_639_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_639_transpose_y_0 = const()[name = tensor<string, []>("op_639_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = tensor<string, []>("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_52 = transpose(perm = transpose_52_perm_0, x = var_630_cast_fp16)[name = tensor<string, []>("transpose_78")];
            tensor<fp16, [1, 12, 512, 64]> transpose_51 = transpose(perm = transpose_51_perm_0, x = var_624_cast_fp16)[name = tensor<string, []>("transpose_79")];
            tensor<fp16, [1, 12, 512, 512]> var_639_cast_fp16 = matmul(transpose_x = var_639_transpose_x_0, transpose_y = var_639_transpose_y_0, x = transpose_51, y = transpose_52)[name = tensor<string, []>("op_639_cast_fp16")];
            tensor<fp16, []> var_640_to_fp16 = const()[name = tensor<string, []>("op_640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_29_cast_fp16 = mul(x = var_639_cast_fp16, y = var_640_to_fp16)[name = tensor<string, []>("attn_weights_29_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_147_cast_fp16 = add(x = attn_weights_29_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_643_cast_fp16 = softmax(axis = var_21, x = input_147_cast_fp16)[name = tensor<string, []>("op_643_cast_fp16")];
            tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = var_636_cast_fp16)[name = tensor<string, []>("transpose_80")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = var_643_cast_fp16, y = value_15_cast_fp16)[name = tensor<string, []>("attn_output_29_cast_fp16")];
            tensor<int32, [4]> var_647_perm_0 = const()[name = tensor<string, []>("op_647_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_649 = const()[name = tensor<string, []>("op_649"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_647_cast_fp16 = transpose(perm = var_647_perm_0, x = attn_output_29_cast_fp16)[name = tensor<string, []>("transpose_77")];
            tensor<fp16, [1, 512, 768]> var_650_cast_fp16 = reshape(shape = var_649, x = var_647_cast_fp16)[name = tensor<string, []>("op_650_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_7_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181831552)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183011264)))];
            tensor<fp16, [1, 512, 768]> linear_45_cast_fp16 = linear(bias = text_model_encoder_layer_7_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_7_attention_output_dense_weight_to_fp16, x = var_650_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_155_cast_fp16 = add(x = linear_45_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183012864)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183014464)))];
            tensor<fp16, [1, 512, 768]> input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_model_encoder_layer_7_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_7_attention_output_LayerNorm_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_7_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183016064)))];
            tensor<fp16, [3072]> text_model_encoder_layer_7_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187734720)))];
            tensor<fp16, [1, 512, 3072]> linear_46_cast_fp16 = linear(bias = text_model_encoder_layer_7_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_7_intermediate_dense_weight_to_fp16, x = input_157_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<string, []> input_161_mode_0 = const()[name = tensor<string, []>("input_161_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_161_cast_fp16 = gelu(mode = input_161_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_7_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187740928)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192459584)))];
            tensor<fp16, [1, 512, 768]> linear_47_cast_fp16 = linear(bias = text_model_encoder_layer_7_output_dense_bias_to_fp16, weight = text_model_encoder_layer_7_output_dense_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_165_cast_fp16 = add(x = linear_47_cast_fp16, y = input_157_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<int32, [1]> hidden_states_47_axes_0 = const()[name = tensor<string, []>("hidden_states_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_7_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192461184)))];
            tensor<fp16, [768]> text_model_encoder_layer_7_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_7_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192462784)))];
            tensor<fp16, [1, 512, 768]> hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = text_model_encoder_layer_7_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_7_output_LayerNorm_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("hidden_states_47_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_8_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192464384)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193644096)))];
            tensor<fp16, [1, 512, 768]> linear_48_cast_fp16 = linear(bias = text_model_encoder_layer_8_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_8_attention_self_query_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<int32, [4]> var_692 = const()[name = tensor<string, []>("op_692"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_693_cast_fp16 = reshape(shape = var_692, x = linear_48_cast_fp16)[name = tensor<string, []>("op_693_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_8_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193645696)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194825408)))];
            tensor<fp16, [1, 512, 768]> linear_49_cast_fp16 = linear(bias = text_model_encoder_layer_8_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_8_attention_self_key_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<int32, [4]> var_698 = const()[name = tensor<string, []>("op_698"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_699_cast_fp16 = reshape(shape = var_698, x = linear_49_cast_fp16)[name = tensor<string, []>("op_699_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_8_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194827008)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196006720)))];
            tensor<fp16, [1, 512, 768]> linear_50_cast_fp16 = linear(bias = text_model_encoder_layer_8_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_8_attention_self_value_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_704 = const()[name = tensor<string, []>("op_704"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_705_cast_fp16 = reshape(shape = var_704, x = linear_50_cast_fp16)[name = tensor<string, []>("op_705_cast_fp16")];
            tensor<int32, [4]> value_17_perm_0 = const()[name = tensor<string, []>("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_708_transpose_x_0 = const()[name = tensor<string, []>("op_708_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_708_transpose_y_0 = const()[name = tensor<string, []>("op_708_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = tensor<string, []>("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = tensor<string, []>("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_54 = transpose(perm = transpose_54_perm_0, x = var_699_cast_fp16)[name = tensor<string, []>("transpose_74")];
            tensor<fp16, [1, 12, 512, 64]> transpose_53 = transpose(perm = transpose_53_perm_0, x = var_693_cast_fp16)[name = tensor<string, []>("transpose_75")];
            tensor<fp16, [1, 12, 512, 512]> var_708_cast_fp16 = matmul(transpose_x = var_708_transpose_x_0, transpose_y = var_708_transpose_y_0, x = transpose_53, y = transpose_54)[name = tensor<string, []>("op_708_cast_fp16")];
            tensor<fp16, []> var_709_to_fp16 = const()[name = tensor<string, []>("op_709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_33_cast_fp16 = mul(x = var_708_cast_fp16, y = var_709_to_fp16)[name = tensor<string, []>("attn_weights_33_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_167_cast_fp16 = add(x = attn_weights_33_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_712_cast_fp16 = softmax(axis = var_21, x = input_167_cast_fp16)[name = tensor<string, []>("op_712_cast_fp16")];
            tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = var_705_cast_fp16)[name = tensor<string, []>("transpose_76")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = var_712_cast_fp16, y = value_17_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")];
            tensor<int32, [4]> var_716_perm_0 = const()[name = tensor<string, []>("op_716_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_718 = const()[name = tensor<string, []>("op_718"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_716_cast_fp16 = transpose(perm = var_716_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_73")];
            tensor<fp16, [1, 512, 768]> var_719_cast_fp16 = reshape(shape = var_718, x = var_716_cast_fp16)[name = tensor<string, []>("op_719_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_8_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196008320)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197188032)))];
            tensor<fp16, [1, 512, 768]> linear_51_cast_fp16 = linear(bias = text_model_encoder_layer_8_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_8_attention_output_dense_weight_to_fp16, x = var_719_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_175_cast_fp16 = add(x = linear_51_cast_fp16, y = hidden_states_47_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<int32, [1]> input_177_axes_0 = const()[name = tensor<string, []>("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197189632)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197191232)))];
            tensor<fp16, [1, 512, 768]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = text_model_encoder_layer_8_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_8_attention_output_LayerNorm_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_8_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197192832)))];
            tensor<fp16, [3072]> text_model_encoder_layer_8_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201911488)))];
            tensor<fp16, [1, 512, 3072]> linear_52_cast_fp16 = linear(bias = text_model_encoder_layer_8_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_8_intermediate_dense_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<string, []> input_181_mode_0 = const()[name = tensor<string, []>("input_181_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_181_cast_fp16 = gelu(mode = input_181_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_8_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201917696)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206636352)))];
            tensor<fp16, [1, 512, 768]> linear_53_cast_fp16 = linear(bias = text_model_encoder_layer_8_output_dense_bias_to_fp16, weight = text_model_encoder_layer_8_output_dense_weight_to_fp16, x = input_181_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_185_cast_fp16 = add(x = linear_53_cast_fp16, y = input_177_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<int32, [1]> hidden_states_53_axes_0 = const()[name = tensor<string, []>("hidden_states_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_8_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206637952)))];
            tensor<fp16, [768]> text_model_encoder_layer_8_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_8_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206639552)))];
            tensor<fp16, [1, 512, 768]> hidden_states_53_cast_fp16 = layer_norm(axes = hidden_states_53_axes_0, beta = text_model_encoder_layer_8_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_8_output_LayerNorm_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("hidden_states_53_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_9_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206641152)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207820864)))];
            tensor<fp16, [1, 512, 768]> linear_54_cast_fp16 = linear(bias = text_model_encoder_layer_9_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_9_attention_self_query_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<int32, [4]> var_761 = const()[name = tensor<string, []>("op_761"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_762_cast_fp16 = reshape(shape = var_761, x = linear_54_cast_fp16)[name = tensor<string, []>("op_762_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_9_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207822464)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209002176)))];
            tensor<fp16, [1, 512, 768]> linear_55_cast_fp16 = linear(bias = text_model_encoder_layer_9_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_9_attention_self_key_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<int32, [4]> var_767 = const()[name = tensor<string, []>("op_767"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_768_cast_fp16 = reshape(shape = var_767, x = linear_55_cast_fp16)[name = tensor<string, []>("op_768_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_9_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209003776)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210183488)))];
            tensor<fp16, [1, 512, 768]> linear_56_cast_fp16 = linear(bias = text_model_encoder_layer_9_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_9_attention_self_value_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_773 = const()[name = tensor<string, []>("op_773"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_774_cast_fp16 = reshape(shape = var_773, x = linear_56_cast_fp16)[name = tensor<string, []>("op_774_cast_fp16")];
            tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_777_transpose_x_0 = const()[name = tensor<string, []>("op_777_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_777_transpose_y_0 = const()[name = tensor<string, []>("op_777_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_55_perm_0 = const()[name = tensor<string, []>("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_56 = transpose(perm = transpose_56_perm_0, x = var_768_cast_fp16)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [1, 12, 512, 64]> transpose_55 = transpose(perm = transpose_55_perm_0, x = var_762_cast_fp16)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [1, 12, 512, 512]> var_777_cast_fp16 = matmul(transpose_x = var_777_transpose_x_0, transpose_y = var_777_transpose_y_0, x = transpose_55, y = transpose_56)[name = tensor<string, []>("op_777_cast_fp16")];
            tensor<fp16, []> var_778_to_fp16 = const()[name = tensor<string, []>("op_778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_37_cast_fp16 = mul(x = var_777_cast_fp16, y = var_778_to_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_187_cast_fp16 = add(x = attn_weights_37_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_781_cast_fp16 = softmax(axis = var_21, x = input_187_cast_fp16)[name = tensor<string, []>("op_781_cast_fp16")];
            tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = var_774_cast_fp16)[name = tensor<string, []>("transpose_72")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = var_781_cast_fp16, y = value_19_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")];
            tensor<int32, [4]> var_785_perm_0 = const()[name = tensor<string, []>("op_785_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_787 = const()[name = tensor<string, []>("op_787"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_785_cast_fp16 = transpose(perm = var_785_perm_0, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [1, 512, 768]> var_788_cast_fp16 = reshape(shape = var_787, x = var_785_cast_fp16)[name = tensor<string, []>("op_788_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_9_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210185088)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211364800)))];
            tensor<fp16, [1, 512, 768]> linear_57_cast_fp16 = linear(bias = text_model_encoder_layer_9_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_9_attention_output_dense_weight_to_fp16, x = var_788_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_195_cast_fp16 = add(x = linear_57_cast_fp16, y = hidden_states_53_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<int32, [1]> input_197_axes_0 = const()[name = tensor<string, []>("input_197_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211366400)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211368000)))];
            tensor<fp16, [1, 512, 768]> input_197_cast_fp16 = layer_norm(axes = input_197_axes_0, beta = text_model_encoder_layer_9_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_9_attention_output_LayerNorm_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_9_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211369600)))];
            tensor<fp16, [3072]> text_model_encoder_layer_9_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216088256)))];
            tensor<fp16, [1, 512, 3072]> linear_58_cast_fp16 = linear(bias = text_model_encoder_layer_9_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_9_intermediate_dense_weight_to_fp16, x = input_197_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<string, []> input_201_mode_0 = const()[name = tensor<string, []>("input_201_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_201_cast_fp16 = gelu(mode = input_201_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_9_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216094464)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220813120)))];
            tensor<fp16, [1, 512, 768]> linear_59_cast_fp16 = linear(bias = text_model_encoder_layer_9_output_dense_bias_to_fp16, weight = text_model_encoder_layer_9_output_dense_weight_to_fp16, x = input_201_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_205_cast_fp16 = add(x = linear_59_cast_fp16, y = input_197_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<int32, [1]> hidden_states_59_axes_0 = const()[name = tensor<string, []>("hidden_states_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_9_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220814720)))];
            tensor<fp16, [768]> text_model_encoder_layer_9_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_9_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220816320)))];
            tensor<fp16, [1, 512, 768]> hidden_states_59_cast_fp16 = layer_norm(axes = hidden_states_59_axes_0, beta = text_model_encoder_layer_9_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_9_output_LayerNorm_weight_to_fp16, x = input_205_cast_fp16)[name = tensor<string, []>("hidden_states_59_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_10_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220817920)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221997632)))];
            tensor<fp16, [1, 512, 768]> linear_60_cast_fp16 = linear(bias = text_model_encoder_layer_10_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_10_attention_self_query_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<int32, [4]> var_830 = const()[name = tensor<string, []>("op_830"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_831_cast_fp16 = reshape(shape = var_830, x = linear_60_cast_fp16)[name = tensor<string, []>("op_831_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_10_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221999232)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223178944)))];
            tensor<fp16, [1, 512, 768]> linear_61_cast_fp16 = linear(bias = text_model_encoder_layer_10_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_10_attention_self_key_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<int32, [4]> var_836 = const()[name = tensor<string, []>("op_836"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_837_cast_fp16 = reshape(shape = var_836, x = linear_61_cast_fp16)[name = tensor<string, []>("op_837_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_10_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223180544)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224360256)))];
            tensor<fp16, [1, 512, 768]> linear_62_cast_fp16 = linear(bias = text_model_encoder_layer_10_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_10_attention_self_value_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_842 = const()[name = tensor<string, []>("op_842"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_843_cast_fp16 = reshape(shape = var_842, x = linear_62_cast_fp16)[name = tensor<string, []>("op_843_cast_fp16")];
            tensor<int32, [4]> value_21_perm_0 = const()[name = tensor<string, []>("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_846_transpose_x_0 = const()[name = tensor<string, []>("op_846_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_846_transpose_y_0 = const()[name = tensor<string, []>("op_846_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = tensor<string, []>("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_58 = transpose(perm = transpose_58_perm_0, x = var_837_cast_fp16)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [1, 12, 512, 64]> transpose_57 = transpose(perm = transpose_57_perm_0, x = var_831_cast_fp16)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [1, 12, 512, 512]> var_846_cast_fp16 = matmul(transpose_x = var_846_transpose_x_0, transpose_y = var_846_transpose_y_0, x = transpose_57, y = transpose_58)[name = tensor<string, []>("op_846_cast_fp16")];
            tensor<fp16, []> var_847_to_fp16 = const()[name = tensor<string, []>("op_847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_41_cast_fp16 = mul(x = var_846_cast_fp16, y = var_847_to_fp16)[name = tensor<string, []>("attn_weights_41_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_207_cast_fp16 = add(x = attn_weights_41_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_850_cast_fp16 = softmax(axis = var_21, x = input_207_cast_fp16)[name = tensor<string, []>("op_850_cast_fp16")];
            tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = var_843_cast_fp16)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = var_850_cast_fp16, y = value_21_cast_fp16)[name = tensor<string, []>("attn_output_41_cast_fp16")];
            tensor<int32, [4]> var_854_perm_0 = const()[name = tensor<string, []>("op_854_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_856 = const()[name = tensor<string, []>("op_856"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_854_cast_fp16 = transpose(perm = var_854_perm_0, x = attn_output_41_cast_fp16)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [1, 512, 768]> var_857_cast_fp16 = reshape(shape = var_856, x = var_854_cast_fp16)[name = tensor<string, []>("op_857_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_10_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224361856)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225541568)))];
            tensor<fp16, [1, 512, 768]> linear_63_cast_fp16 = linear(bias = text_model_encoder_layer_10_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_10_attention_output_dense_weight_to_fp16, x = var_857_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_215_cast_fp16 = add(x = linear_63_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<int32, [1]> input_217_axes_0 = const()[name = tensor<string, []>("input_217_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225543168)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225544768)))];
            tensor<fp16, [1, 512, 768]> input_217_cast_fp16 = layer_norm(axes = input_217_axes_0, beta = text_model_encoder_layer_10_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_10_attention_output_LayerNorm_weight_to_fp16, x = input_215_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_10_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225546368)))];
            tensor<fp16, [3072]> text_model_encoder_layer_10_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230265024)))];
            tensor<fp16, [1, 512, 3072]> linear_64_cast_fp16 = linear(bias = text_model_encoder_layer_10_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_10_intermediate_dense_weight_to_fp16, x = input_217_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<string, []> input_221_mode_0 = const()[name = tensor<string, []>("input_221_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_221_cast_fp16 = gelu(mode = input_221_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_10_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230271232)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234989888)))];
            tensor<fp16, [1, 512, 768]> linear_65_cast_fp16 = linear(bias = text_model_encoder_layer_10_output_dense_bias_to_fp16, weight = text_model_encoder_layer_10_output_dense_weight_to_fp16, x = input_221_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_225_cast_fp16 = add(x = linear_65_cast_fp16, y = input_217_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
            tensor<int32, [1]> hidden_states_65_axes_0 = const()[name = tensor<string, []>("hidden_states_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_10_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234991488)))];
            tensor<fp16, [768]> text_model_encoder_layer_10_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_10_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234993088)))];
            tensor<fp16, [1, 512, 768]> hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = text_model_encoder_layer_10_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_10_output_LayerNorm_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("hidden_states_65_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_11_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234994688)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236174400)))];
            tensor<fp16, [1, 512, 768]> linear_66_cast_fp16 = linear(bias = text_model_encoder_layer_11_attention_self_query_bias_to_fp16, weight = text_model_encoder_layer_11_attention_self_query_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<int32, [4]> var_899 = const()[name = tensor<string, []>("op_899"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_900_cast_fp16 = reshape(shape = var_899, x = linear_66_cast_fp16)[name = tensor<string, []>("op_900_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_11_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236176000)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237355712)))];
            tensor<fp16, [1, 512, 768]> linear_67_cast_fp16 = linear(bias = text_model_encoder_layer_11_attention_self_key_bias_to_fp16, weight = text_model_encoder_layer_11_attention_self_key_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<int32, [4]> var_905 = const()[name = tensor<string, []>("op_905"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_906_cast_fp16 = reshape(shape = var_905, x = linear_67_cast_fp16)[name = tensor<string, []>("op_906_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_11_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237357312)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238537024)))];
            tensor<fp16, [1, 512, 768]> linear_68_cast_fp16 = linear(bias = text_model_encoder_layer_11_attention_self_value_bias_to_fp16, weight = text_model_encoder_layer_11_attention_self_value_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_911 = const()[name = tensor<string, []>("op_911"), val = tensor<int32, [4]>([1, 512, -1, 64])];
            tensor<fp16, [1, 512, 12, 64]> var_912_cast_fp16 = reshape(shape = var_911, x = linear_68_cast_fp16)[name = tensor<string, []>("op_912_cast_fp16")];
            tensor<int32, [4]> value_23_perm_0 = const()[name = tensor<string, []>("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_915_transpose_x_0 = const()[name = tensor<string, []>("op_915_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_915_transpose_y_0 = const()[name = tensor<string, []>("op_915_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_60_perm_0 = const()[name = tensor<string, []>("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 12, 64, 512]> transpose_60 = transpose(perm = transpose_60_perm_0, x = var_906_cast_fp16)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [1, 12, 512, 64]> transpose_59 = transpose(perm = transpose_59_perm_0, x = var_900_cast_fp16)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [1, 12, 512, 512]> var_915_cast_fp16 = matmul(transpose_x = var_915_transpose_x_0, transpose_y = var_915_transpose_y_0, x = transpose_59, y = transpose_60)[name = tensor<string, []>("op_915_cast_fp16")];
            tensor<fp16, []> var_916_to_fp16 = const()[name = tensor<string, []>("op_916_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 12, 512, 512]> attn_weights_45_cast_fp16 = mul(x = var_915_cast_fp16, y = var_916_to_fp16)[name = tensor<string, []>("attn_weights_45_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> input_227_cast_fp16 = add(x = attn_weights_45_cast_fp16, y = attention_mask_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<fp16, [1, 12, 512, 512]> var_919_cast_fp16 = softmax(axis = var_21, x = input_227_cast_fp16)[name = tensor<string, []>("op_919_cast_fp16")];
            tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 12, 512, 64]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = var_912_cast_fp16)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [1, 12, 512, 64]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = var_919_cast_fp16, y = value_23_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")];
            tensor<int32, [4]> var_923_perm_0 = const()[name = tensor<string, []>("op_923_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_925 = const()[name = tensor<string, []>("op_925"), val = tensor<int32, [3]>([1, 512, -1])];
            tensor<fp16, [1, 512, 12, 64]> var_923_cast_fp16 = transpose(perm = var_923_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [1, 512, 768]> var_926_cast_fp16 = reshape(shape = var_925, x = var_923_cast_fp16)[name = tensor<string, []>("op_926_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_encoder_layer_11_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238538624)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239718336)))];
            tensor<fp16, [1, 512, 768]> linear_69_cast_fp16 = linear(bias = text_model_encoder_layer_11_attention_output_dense_bias_to_fp16, weight = text_model_encoder_layer_11_attention_output_dense_weight_to_fp16, x = var_926_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_235_cast_fp16 = add(x = linear_69_cast_fp16, y = hidden_states_65_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")];
            tensor<int32, [1]> input_237_axes_0 = const()[name = tensor<string, []>("input_237_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239719936)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239721536)))];
            tensor<fp16, [1, 512, 768]> input_237_cast_fp16 = layer_norm(axes = input_237_axes_0, beta = text_model_encoder_layer_11_attention_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_11_attention_output_LayerNorm_weight_to_fp16, x = input_235_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")];
            tensor<fp16, [3072, 768]> text_model_encoder_layer_11_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239723136)))];
            tensor<fp16, [3072]> text_model_encoder_layer_11_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244441792)))];
            tensor<fp16, [1, 512, 3072]> linear_70_cast_fp16 = linear(bias = text_model_encoder_layer_11_intermediate_dense_bias_to_fp16, weight = text_model_encoder_layer_11_intermediate_dense_weight_to_fp16, x = input_237_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<string, []> input_241_mode_0 = const()[name = tensor<string, []>("input_241_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 512, 3072]> input_241_cast_fp16 = gelu(mode = input_241_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<fp16, [768, 3072]> text_model_encoder_layer_11_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244448000)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249166656)))];
            tensor<fp16, [1, 512, 768]> linear_71_cast_fp16 = linear(bias = text_model_encoder_layer_11_output_dense_bias_to_fp16, weight = text_model_encoder_layer_11_output_dense_weight_to_fp16, x = input_241_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 512, 768]> input_245_cast_fp16 = add(x = linear_71_cast_fp16, y = input_237_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")];
            tensor<int32, [1]> hidden_states_axes_0 = const()[name = tensor<string, []>("hidden_states_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> text_model_encoder_layer_11_output_LayerNorm_weight_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_output_LayerNorm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249168256)))];
            tensor<fp16, [768]> text_model_encoder_layer_11_output_LayerNorm_bias_to_fp16 = const()[name = tensor<string, []>("text_model_encoder_layer_11_output_LayerNorm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249169856)))];
            tensor<fp16, [1, 512, 768]> hidden_states_cast_fp16 = layer_norm(axes = hidden_states_axes_0, beta = text_model_encoder_layer_11_output_LayerNorm_bias_to_fp16, epsilon = var_23_to_fp16, gamma = text_model_encoder_layer_11_output_LayerNorm_weight_to_fp16, x = input_245_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")];
            tensor<int32, [3]> input_247_begin_0 = const()[name = tensor<string, []>("input_247_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> input_247_end_0 = const()[name = tensor<string, []>("input_247_end_0"), val = tensor<int32, [3]>([1, 1, 768])];
            tensor<bool, [3]> input_247_end_mask_0 = const()[name = tensor<string, []>("input_247_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> input_247_squeeze_mask_0 = const()[name = tensor<string, []>("input_247_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<fp16, [1, 768]> input_247_cast_fp16 = slice_by_index(begin = input_247_begin_0, end = input_247_end_0, end_mask = input_247_end_mask_0, squeeze_mask = input_247_squeeze_mask_0, x = hidden_states_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")];
            tensor<fp16, [768, 768]> text_model_pooler_dense_weight_to_fp16 = const()[name = tensor<string, []>("text_model_pooler_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249171456)))];
            tensor<fp16, [768]> text_model_pooler_dense_bias_to_fp16 = const()[name = tensor<string, []>("text_model_pooler_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250351168)))];
            tensor<fp16, [1, 768]> linear_72_cast_fp16 = linear(bias = text_model_pooler_dense_bias_to_fp16, weight = text_model_pooler_dense_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1, 768]> input_251_cast_fp16 = tanh(x = linear_72_cast_fp16)[name = tensor<string, []>("input_251_cast_fp16")];
            tensor<fp16, [512, 768]> text_projection_linear1_weight_to_fp16 = const()[name = tensor<string, []>("text_projection_linear1_weight_to_fp16"), val = tensor<fp16, [512, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250352768)))];
            tensor<fp16, [512]> text_projection_linear1_bias_to_fp16 = const()[name = tensor<string, []>("text_projection_linear1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251139264)))];
            tensor<fp16, [1, 512]> linear_73_cast_fp16 = linear(bias = text_projection_linear1_bias_to_fp16, weight = text_projection_linear1_weight_to_fp16, x = input_251_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1, 512]> input_cast_fp16 = relu(x = linear_73_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<fp16, [512, 512]> text_projection_linear2_weight_to_fp16 = const()[name = tensor<string, []>("text_projection_linear2_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251140352)))];
            tensor<fp16, [512]> text_projection_linear2_bias_to_fp16 = const()[name = tensor<string, []>("text_projection_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251664704)))];
            tensor<fp16, [1, 512]> linear_74_cast_fp16 = linear(bias = text_projection_linear2_bias_to_fp16, weight = text_projection_linear2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<string, []> linear_74_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_74_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 512]> text_embeds = cast(dtype = linear_74_cast_fp16_to_fp32_dtype_0, x = linear_74_cast_fp16)[name = tensor<string, []>("cast_56")];
        } -> (text_embeds);
}