program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.10.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] { func main(tensor attention_mask, tensor input_ids, tensor span_idx, tensor span_mask, tensor text_lengths, tensor words_mask) { tensor word_positions = const()[name = string("word_positions"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; int32 var_32 = const()[name = string("op_32"), val = int32(-1)]; int32 inputs_embeds_axis_0 = const()[name = string("inputs_embeds_axis_0"), val = int32(0)]; int32 inputs_embeds_batch_dims_0 = const()[name = string("inputs_embeds_batch_dims_0"), val = int32(0)]; bool inputs_embeds_validate_indices_0 = const()[name = string("inputs_embeds_validate_indices_0"), val = bool(false)]; tensor deberta_embeddings_word_embeddings_weight_to_fp16 = const()[name = string("deberta_embeddings_word_embeddings_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1408)))]; tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = input_ids, validate_indices = inputs_embeds_validate_indices_0, x = deberta_embeddings_word_embeddings_weight_to_fp16)[name = string("inputs_embeds_cast_fp16")]; tensor embeddings_1_axes_0 = const()[name = string("embeddings_1_axes_0"), val = tensor([-1])]; tensor deberta_embeddings_LayerNorm_weight_to_fp16 = const()[name = string("deberta_embeddings_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196615616)))]; tensor deberta_embeddings_LayerNorm_bias_to_fp16 = const()[name = string("deberta_embeddings_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196617216)))]; fp16 var_51_to_fp16 = const()[name = string("op_51_to_fp16"), val = fp16(0x1p-23)]; tensor embeddings_1_cast_fp16 = layer_norm(axes = embeddings_1_axes_0, beta = deberta_embeddings_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_embeddings_LayerNorm_weight_to_fp16, x = inputs_embeds_cast_fp16)[name = string("embeddings_1_cast_fp16")]; tensor mask_3_axes_0 = const()[name = string("mask_3_axes_0"), val = tensor([2])]; tensor mask_3 = expand_dims(axes = mask_3_axes_0, x = attention_mask)[name = string("mask_3")]; string cast_3_to_fp16_dtype_0 = const()[name = string("cast_3_to_fp16_dtype_0"), val = string("fp16")]; tensor mask_3_to_fp16 = cast(dtype = cast_3_to_fp16_dtype_0, x = mask_3)[name = string("cast_295")]; tensor input_1_cast_fp16 = mul(x = embeddings_1_cast_fp16, y = mask_3_to_fp16)[name = string("input_1_cast_fp16")]; tensor var_98_axes_0 = const()[name = string("op_98_axes_0"), val = tensor([1])]; tensor var_98 = expand_dims(axes = var_98_axes_0, x = attention_mask)[name = string("op_98")]; tensor extended_attention_mask_axes_0 = const()[name = string("extended_attention_mask_axes_0"), val = tensor([2])]; tensor extended_attention_mask = expand_dims(axes = extended_attention_mask_axes_0, x = var_98)[name = string("extended_attention_mask")]; tensor var_100_axes_0 = const()[name = string("op_100_axes_0"), val = tensor([-2])]; string extended_attention_mask_to_uint16_dtype_0 = const()[name = string("extended_attention_mask_to_uint16_dtype_0"), val = string("uint16")]; tensor extended_attention_mask_to_uint16 = cast(dtype = extended_attention_mask_to_uint16_dtype_0, x = extended_attention_mask)[name = string("cast_294")]; tensor var_100_cast_int16 = squeeze(axes = var_100_axes_0, x = extended_attention_mask_to_uint16)[name = string("op_100_cast_int16")]; string var_100_cast_int16_to_int32_dtype_0 = const()[name = string("op_100_cast_int16_to_int32_dtype_0"), val = string("int32")]; tensor var_101_axes_0 = const()[name = string("op_101_axes_0"), val = tensor([-1])]; tensor var_100_cast_int16_to_int32 = cast(dtype = var_100_cast_int16_to_int32_dtype_0, x = var_100_cast_int16)[name = string("cast_293")]; tensor var_101 = expand_dims(axes = var_101_axes_0, x = var_100_cast_int16_to_int32)[name = string("op_101")]; tensor attention_mask_3 = mul(x = extended_attention_mask, y = var_101)[name = string("attention_mask_3")]; tensor deberta_encoder_layer_0_attention_self_query_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_query_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196618816)))]; tensor deberta_encoder_layer_0_attention_self_query_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_query_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197798528)))]; tensor linear_0_cast_fp16 = linear(bias = deberta_encoder_layer_0_attention_self_query_proj_bias_to_fp16, weight = deberta_encoder_layer_0_attention_self_query_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_0_cast_fp16")]; tensor var_158 = const()[name = string("op_158"), val = tensor([1, 512, 12, -1])]; tensor x_3_cast_fp16 = reshape(shape = var_158, x = linear_0_cast_fp16)[name = string("x_3_cast_fp16")]; tensor var_160 = const()[name = string("op_160"), val = tensor([0, 2, 1, 3])]; tensor var_165 = const()[name = string("op_165"), val = tensor([-1, 512, 64])]; tensor var_161_cast_fp16 = transpose(perm = var_160, x = x_3_cast_fp16)[name = string("transpose_79")]; tensor query_layer_3_cast_fp16 = reshape(shape = var_165, x = var_161_cast_fp16)[name = string("query_layer_3_cast_fp16")]; tensor deberta_encoder_layer_0_attention_self_key_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_key_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197800128)))]; tensor deberta_encoder_layer_0_attention_self_key_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_key_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198979840)))]; tensor linear_1_cast_fp16 = linear(bias = deberta_encoder_layer_0_attention_self_key_proj_bias_to_fp16, weight = deberta_encoder_layer_0_attention_self_key_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_1_cast_fp16")]; tensor var_172 = const()[name = string("op_172"), val = tensor([1, 512, 12, -1])]; tensor x_7_cast_fp16 = reshape(shape = var_172, x = linear_1_cast_fp16)[name = string("x_7_cast_fp16")]; tensor var_174 = const()[name = string("op_174"), val = tensor([0, 2, 1, 3])]; tensor var_179 = const()[name = string("op_179"), val = tensor([-1, 512, 64])]; tensor var_175_cast_fp16 = transpose(perm = var_174, x = x_7_cast_fp16)[name = string("transpose_78")]; tensor key_layer_1_cast_fp16 = reshape(shape = var_179, x = var_175_cast_fp16)[name = string("key_layer_1_cast_fp16")]; tensor deberta_encoder_layer_0_attention_self_value_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_value_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198981440)))]; tensor deberta_encoder_layer_0_attention_self_value_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_self_value_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200161152)))]; tensor linear_2_cast_fp16 = linear(bias = deberta_encoder_layer_0_attention_self_value_proj_bias_to_fp16, weight = deberta_encoder_layer_0_attention_self_value_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_2_cast_fp16")]; tensor var_186 = const()[name = string("op_186"), val = tensor([1, 512, 12, -1])]; tensor x_11_cast_fp16 = reshape(shape = var_186, x = linear_2_cast_fp16)[name = string("x_11_cast_fp16")]; tensor var_188 = const()[name = string("op_188"), val = tensor([0, 2, 1, 3])]; tensor var_193 = const()[name = string("op_193"), val = tensor([-1, 512, 64])]; tensor var_189_cast_fp16 = transpose(perm = var_188, x = x_11_cast_fp16)[name = string("transpose_77")]; tensor value_layer_1_cast_fp16 = reshape(shape = var_193, x = var_189_cast_fp16)[name = string("value_layer_1_cast_fp16")]; tensor var_203_perm_0 = const()[name = string("op_203_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_205_y_0_to_fp16 = const()[name = string("_inversed_205_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor var_203_cast_fp16 = transpose(perm = var_203_perm_0, x = key_layer_1_cast_fp16)[name = string("transpose_76")]; tensor _inversed_205_cast_fp16 = mul(x = var_203_cast_fp16, y = _inversed_205_y_0_to_fp16)[name = string("_inversed_205_cast_fp16")]; bool attention_scores_1_transpose_x_0 = const()[name = string("attention_scores_1_transpose_x_0"), val = bool(false)]; bool attention_scores_1_transpose_y_0 = const()[name = string("attention_scores_1_transpose_y_0"), val = bool(false)]; tensor attention_scores_1_cast_fp16 = matmul(transpose_x = attention_scores_1_transpose_x_0, transpose_y = attention_scores_1_transpose_y_0, x = query_layer_3_cast_fp16, y = _inversed_205_cast_fp16)[name = string("attention_scores_1_cast_fp16")]; bool c2p_att_1_transpose_x_0 = const()[name = string("c2p_att_1_transpose_x_0"), val = bool(false)]; bool c2p_att_1_transpose_y_0 = const()[name = string("c2p_att_1_transpose_y_0"), val = bool(false)]; tensor var_261_to_fp16 = const()[name = string("op_261_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200162752)))]; tensor c2p_att_1_cast_fp16 = matmul(transpose_x = c2p_att_1_transpose_x_0, transpose_y = c2p_att_1_transpose_y_0, x = query_layer_3_cast_fp16, y = var_261_to_fp16)[name = string("c2p_att_1_cast_fp16")]; bool c2p_att_3_validate_indices_0 = const()[name = string("c2p_att_3_validate_indices_0"), val = bool(false)]; tensor var_271_to_uint16 = const()[name = string("op_271_to_uint16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200949248)))]; tensor c2p_att_3_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_271_to_uint16, validate_indices = c2p_att_3_validate_indices_0, x = c2p_att_1_cast_fp16)[name = string("c2p_att_3_cast_fp16_cast_int16")]; tensor _inversed_274_y_0_to_fp16 = const()[name = string("_inversed_274_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor _inversed_274_cast_fp16 = mul(x = c2p_att_3_cast_fp16_cast_int16, y = _inversed_274_y_0_to_fp16)[name = string("_inversed_274_cast_fp16")]; bool p2c_att_1_transpose_x_0 = const()[name = string("p2c_att_1_transpose_x_0"), val = bool(false)]; bool p2c_att_1_transpose_y_0 = const()[name = string("p2c_att_1_transpose_y_0"), val = bool(false)]; tensor var_289_to_fp16 = const()[name = string("op_289_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207240768)))]; tensor p2c_att_1_cast_fp16 = matmul(transpose_x = p2c_att_1_transpose_x_0, transpose_y = p2c_att_1_transpose_y_0, x = key_layer_1_cast_fp16, y = var_289_to_fp16)[name = string("p2c_att_1_cast_fp16")]; bool var_297_validate_indices_0 = const()[name = string("op_297_validate_indices_0"), val = bool(false)]; tensor var_296_to_uint16 = const()[name = string("op_296_to_uint16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208027264)))]; tensor var_297_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_296_to_uint16, validate_indices = var_297_validate_indices_0, x = p2c_att_1_cast_fp16)[name = string("op_297_cast_fp16_cast_int16")]; tensor p2c_att_3_perm_0 = const()[name = string("p2c_att_3_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_300_y_0_to_fp16 = const()[name = string("_inversed_300_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor p2c_att_3_cast_fp16 = transpose(perm = p2c_att_3_perm_0, x = var_297_cast_fp16_cast_int16)[name = string("transpose_75")]; tensor _inversed_300_cast_fp16 = mul(x = p2c_att_3_cast_fp16, y = _inversed_300_y_0_to_fp16)[name = string("_inversed_300_cast_fp16")]; tensor rel_att_1_cast_fp16 = add(x = _inversed_274_cast_fp16, y = _inversed_300_cast_fp16)[name = string("rel_att_1_cast_fp16")]; tensor attention_scores_3_cast_fp16 = add(x = attention_scores_1_cast_fp16, y = rel_att_1_cast_fp16)[name = string("attention_scores_3_cast_fp16")]; tensor var_305 = const()[name = string("op_305"), val = tensor([-1, 12, 512, 512])]; tensor attention_scores_5_cast_fp16 = reshape(shape = var_305, x = attention_scores_3_cast_fp16)[name = string("attention_scores_5_cast_fp16")]; string cast_27_dtype_0 = const()[name = string("cast_27_dtype_0"), val = string("bool")]; tensor cast_27 = cast(dtype = cast_27_dtype_0, x = attention_mask_3)[name = string("cast_292")]; tensor var_308 = logical_not(x = cast_27)[name = string("op_308")]; fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(-inf)]; tensor input_9_cast_fp16 = select(a = var_41_to_fp16, b = attention_scores_5_cast_fp16, cond = var_308)[name = string("input_9_cast_fp16")]; tensor input_11_cast_fp16 = softmax(axis = var_32, x = input_9_cast_fp16)[name = string("input_11_cast_fp16")]; tensor var_314 = const()[name = string("op_314"), val = tensor([-1, 512, 512])]; tensor var_315_cast_fp16 = reshape(shape = var_314, x = input_11_cast_fp16)[name = string("op_315_cast_fp16")]; bool context_layer_1_transpose_x_0 = const()[name = string("context_layer_1_transpose_x_0"), val = bool(false)]; bool context_layer_1_transpose_y_0 = const()[name = string("context_layer_1_transpose_y_0"), val = bool(false)]; tensor context_layer_1_cast_fp16 = matmul(transpose_x = context_layer_1_transpose_x_0, transpose_y = context_layer_1_transpose_y_0, x = var_315_cast_fp16, y = value_layer_1_cast_fp16)[name = string("context_layer_1_cast_fp16")]; tensor var_319 = const()[name = string("op_319"), val = tensor([-1, 12, 512, 64])]; tensor var_320_cast_fp16 = reshape(shape = var_319, x = context_layer_1_cast_fp16)[name = string("op_320_cast_fp16")]; tensor var_321 = const()[name = string("op_321"), val = tensor([0, 2, 1, 3])]; tensor var_326 = const()[name = string("op_326"), val = tensor([1, 512, -1])]; tensor var_322_cast_fp16 = transpose(perm = var_321, x = var_320_cast_fp16)[name = string("transpose_74")]; tensor input_13_cast_fp16 = reshape(shape = var_326, x = var_322_cast_fp16)[name = string("input_13_cast_fp16")]; tensor deberta_encoder_layer_0_attention_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214318784)))]; tensor deberta_encoder_layer_0_attention_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215498496)))]; tensor linear_5_cast_fp16 = linear(bias = deberta_encoder_layer_0_attention_output_dense_bias_to_fp16, weight = deberta_encoder_layer_0_attention_output_dense_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_5_cast_fp16")]; tensor input_17_cast_fp16 = add(x = linear_5_cast_fp16, y = input_1_cast_fp16)[name = string("input_17_cast_fp16")]; tensor input_19_axes_0 = const()[name = string("input_19_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215500096)))]; tensor deberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215501696)))]; tensor input_19_cast_fp16 = layer_norm(axes = input_19_axes_0, beta = deberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; tensor deberta_encoder_layer_0_intermediate_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215503296)))]; tensor deberta_encoder_layer_0_intermediate_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220221952)))]; tensor linear_6_cast_fp16 = linear(bias = deberta_encoder_layer_0_intermediate_dense_bias_to_fp16, weight = deberta_encoder_layer_0_intermediate_dense_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_6_cast_fp16")]; string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = linear_6_cast_fp16)[name = string("input_23_cast_fp16")]; tensor deberta_encoder_layer_0_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220228160)))]; tensor deberta_encoder_layer_0_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224946816)))]; tensor linear_7_cast_fp16 = linear(bias = deberta_encoder_layer_0_output_dense_bias_to_fp16, weight = deberta_encoder_layer_0_output_dense_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_7_cast_fp16")]; tensor input_27_cast_fp16 = add(x = linear_7_cast_fp16, y = input_19_cast_fp16)[name = string("input_27_cast_fp16")]; tensor input_29_axes_0 = const()[name = string("input_29_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_0_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_0_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224948416)))]; tensor deberta_encoder_layer_0_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_0_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224950016)))]; tensor input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = deberta_encoder_layer_0_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_0_output_LayerNorm_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; tensor deberta_encoder_layer_1_attention_self_query_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_query_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224951616)))]; tensor deberta_encoder_layer_1_attention_self_query_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_query_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226131328)))]; tensor linear_8_cast_fp16 = linear(bias = deberta_encoder_layer_1_attention_self_query_proj_bias_to_fp16, weight = deberta_encoder_layer_1_attention_self_query_proj_weight_to_fp16, x = input_29_cast_fp16)[name = string("linear_8_cast_fp16")]; tensor var_377 = const()[name = string("op_377"), val = tensor([1, 512, 12, -1])]; tensor x_23_cast_fp16 = reshape(shape = var_377, x = linear_8_cast_fp16)[name = string("x_23_cast_fp16")]; tensor var_379 = const()[name = string("op_379"), val = tensor([0, 2, 1, 3])]; tensor var_384 = const()[name = string("op_384"), val = tensor([-1, 512, 64])]; tensor var_380_cast_fp16 = transpose(perm = var_379, x = x_23_cast_fp16)[name = string("transpose_73")]; tensor query_layer_9_cast_fp16 = reshape(shape = var_384, x = var_380_cast_fp16)[name = string("query_layer_9_cast_fp16")]; tensor deberta_encoder_layer_1_attention_self_key_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_key_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226132928)))]; tensor deberta_encoder_layer_1_attention_self_key_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_key_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227312640)))]; tensor linear_9_cast_fp16 = linear(bias = deberta_encoder_layer_1_attention_self_key_proj_bias_to_fp16, weight = deberta_encoder_layer_1_attention_self_key_proj_weight_to_fp16, x = input_29_cast_fp16)[name = string("linear_9_cast_fp16")]; tensor var_391 = const()[name = string("op_391"), val = tensor([1, 512, 12, -1])]; tensor x_27_cast_fp16 = reshape(shape = var_391, x = linear_9_cast_fp16)[name = string("x_27_cast_fp16")]; tensor var_393 = const()[name = string("op_393"), val = tensor([0, 2, 1, 3])]; tensor var_398 = const()[name = string("op_398"), val = tensor([-1, 512, 64])]; tensor var_394_cast_fp16 = transpose(perm = var_393, x = x_27_cast_fp16)[name = string("transpose_72")]; tensor key_layer_3_cast_fp16 = reshape(shape = var_398, x = var_394_cast_fp16)[name = string("key_layer_3_cast_fp16")]; tensor deberta_encoder_layer_1_attention_self_value_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_value_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227314240)))]; tensor deberta_encoder_layer_1_attention_self_value_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_self_value_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228493952)))]; tensor linear_10_cast_fp16 = linear(bias = deberta_encoder_layer_1_attention_self_value_proj_bias_to_fp16, weight = deberta_encoder_layer_1_attention_self_value_proj_weight_to_fp16, x = input_29_cast_fp16)[name = string("linear_10_cast_fp16")]; tensor var_405 = const()[name = string("op_405"), val = tensor([1, 512, 12, -1])]; tensor x_31_cast_fp16 = reshape(shape = var_405, x = linear_10_cast_fp16)[name = string("x_31_cast_fp16")]; tensor var_407 = const()[name = string("op_407"), val = tensor([0, 2, 1, 3])]; tensor var_412 = const()[name = string("op_412"), val = tensor([-1, 512, 64])]; tensor var_408_cast_fp16 = transpose(perm = var_407, x = x_31_cast_fp16)[name = string("transpose_71")]; tensor value_layer_3_cast_fp16 = reshape(shape = var_412, x = var_408_cast_fp16)[name = string("value_layer_3_cast_fp16")]; tensor var_422_perm_0 = const()[name = string("op_422_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_424_y_0_to_fp16 = const()[name = string("_inversed_424_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor var_422_cast_fp16 = transpose(perm = var_422_perm_0, x = key_layer_3_cast_fp16)[name = string("transpose_70")]; tensor _inversed_424_cast_fp16 = mul(x = var_422_cast_fp16, y = _inversed_424_y_0_to_fp16)[name = string("_inversed_424_cast_fp16")]; bool attention_scores_7_transpose_x_0 = const()[name = string("attention_scores_7_transpose_x_0"), val = bool(false)]; bool attention_scores_7_transpose_y_0 = const()[name = string("attention_scores_7_transpose_y_0"), val = bool(false)]; tensor attention_scores_7_cast_fp16 = matmul(transpose_x = attention_scores_7_transpose_x_0, transpose_y = attention_scores_7_transpose_y_0, x = query_layer_9_cast_fp16, y = _inversed_424_cast_fp16)[name = string("attention_scores_7_cast_fp16")]; bool c2p_att_5_transpose_x_0 = const()[name = string("c2p_att_5_transpose_x_0"), val = bool(false)]; bool c2p_att_5_transpose_y_0 = const()[name = string("c2p_att_5_transpose_y_0"), val = bool(false)]; tensor var_480_to_fp16 = const()[name = string("op_480_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228495552)))]; tensor c2p_att_5_cast_fp16 = matmul(transpose_x = c2p_att_5_transpose_x_0, transpose_y = c2p_att_5_transpose_y_0, x = query_layer_9_cast_fp16, y = var_480_to_fp16)[name = string("c2p_att_5_cast_fp16")]; bool c2p_att_7_validate_indices_0 = const()[name = string("c2p_att_7_validate_indices_0"), val = bool(false)]; tensor c2p_att_7_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_271_to_uint16, validate_indices = c2p_att_7_validate_indices_0, x = c2p_att_5_cast_fp16)[name = string("c2p_att_7_cast_fp16_cast_int16")]; tensor _inversed_493_y_0_to_fp16 = const()[name = string("_inversed_493_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor _inversed_493_cast_fp16 = mul(x = c2p_att_7_cast_fp16_cast_int16, y = _inversed_493_y_0_to_fp16)[name = string("_inversed_493_cast_fp16")]; bool p2c_att_5_transpose_x_0 = const()[name = string("p2c_att_5_transpose_x_0"), val = bool(false)]; bool p2c_att_5_transpose_y_0 = const()[name = string("p2c_att_5_transpose_y_0"), val = bool(false)]; tensor var_508_to_fp16 = const()[name = string("op_508_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229282048)))]; tensor p2c_att_5_cast_fp16 = matmul(transpose_x = p2c_att_5_transpose_x_0, transpose_y = p2c_att_5_transpose_y_0, x = key_layer_3_cast_fp16, y = var_508_to_fp16)[name = string("p2c_att_5_cast_fp16")]; bool var_516_validate_indices_0 = const()[name = string("op_516_validate_indices_0"), val = bool(false)]; tensor var_516_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_296_to_uint16, validate_indices = var_516_validate_indices_0, x = p2c_att_5_cast_fp16)[name = string("op_516_cast_fp16_cast_int16")]; tensor p2c_att_7_perm_0 = const()[name = string("p2c_att_7_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_519_y_0_to_fp16 = const()[name = string("_inversed_519_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor p2c_att_7_cast_fp16 = transpose(perm = p2c_att_7_perm_0, x = var_516_cast_fp16_cast_int16)[name = string("transpose_69")]; tensor _inversed_519_cast_fp16 = mul(x = p2c_att_7_cast_fp16, y = _inversed_519_y_0_to_fp16)[name = string("_inversed_519_cast_fp16")]; tensor rel_att_3_cast_fp16 = add(x = _inversed_493_cast_fp16, y = _inversed_519_cast_fp16)[name = string("rel_att_3_cast_fp16")]; tensor attention_scores_9_cast_fp16 = add(x = attention_scores_7_cast_fp16, y = rel_att_3_cast_fp16)[name = string("attention_scores_9_cast_fp16")]; tensor var_524 = const()[name = string("op_524"), val = tensor([-1, 12, 512, 512])]; tensor attention_scores_11_cast_fp16 = reshape(shape = var_524, x = attention_scores_9_cast_fp16)[name = string("attention_scores_11_cast_fp16")]; tensor input_33_cast_fp16 = select(a = var_41_to_fp16, b = attention_scores_11_cast_fp16, cond = var_308)[name = string("input_33_cast_fp16")]; tensor input_35_cast_fp16 = softmax(axis = var_32, x = input_33_cast_fp16)[name = string("input_35_cast_fp16")]; tensor var_533 = const()[name = string("op_533"), val = tensor([-1, 512, 512])]; tensor var_534_cast_fp16 = reshape(shape = var_533, x = input_35_cast_fp16)[name = string("op_534_cast_fp16")]; bool context_layer_5_transpose_x_0 = const()[name = string("context_layer_5_transpose_x_0"), val = bool(false)]; bool context_layer_5_transpose_y_0 = const()[name = string("context_layer_5_transpose_y_0"), val = bool(false)]; tensor context_layer_5_cast_fp16 = matmul(transpose_x = context_layer_5_transpose_x_0, transpose_y = context_layer_5_transpose_y_0, x = var_534_cast_fp16, y = value_layer_3_cast_fp16)[name = string("context_layer_5_cast_fp16")]; tensor var_538 = const()[name = string("op_538"), val = tensor([-1, 12, 512, 64])]; tensor var_539_cast_fp16 = reshape(shape = var_538, x = context_layer_5_cast_fp16)[name = string("op_539_cast_fp16")]; tensor var_540 = const()[name = string("op_540"), val = tensor([0, 2, 1, 3])]; tensor var_545 = const()[name = string("op_545"), val = tensor([1, 512, -1])]; tensor var_541_cast_fp16 = transpose(perm = var_540, x = var_539_cast_fp16)[name = string("transpose_68")]; tensor input_37_cast_fp16 = reshape(shape = var_545, x = var_541_cast_fp16)[name = string("input_37_cast_fp16")]; tensor deberta_encoder_layer_1_attention_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230068544)))]; tensor deberta_encoder_layer_1_attention_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231248256)))]; tensor linear_13_cast_fp16 = linear(bias = deberta_encoder_layer_1_attention_output_dense_bias_to_fp16, weight = deberta_encoder_layer_1_attention_output_dense_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_13_cast_fp16")]; tensor input_41_cast_fp16 = add(x = linear_13_cast_fp16, y = input_29_cast_fp16)[name = string("input_41_cast_fp16")]; tensor input_43_axes_0 = const()[name = string("input_43_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231249856)))]; tensor deberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231251456)))]; tensor input_43_cast_fp16 = layer_norm(axes = input_43_axes_0, beta = deberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")]; tensor deberta_encoder_layer_1_intermediate_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231253056)))]; tensor deberta_encoder_layer_1_intermediate_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235971712)))]; tensor linear_14_cast_fp16 = linear(bias = deberta_encoder_layer_1_intermediate_dense_bias_to_fp16, weight = deberta_encoder_layer_1_intermediate_dense_weight_to_fp16, x = input_43_cast_fp16)[name = string("linear_14_cast_fp16")]; string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = linear_14_cast_fp16)[name = string("input_47_cast_fp16")]; tensor deberta_encoder_layer_1_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235977920)))]; tensor deberta_encoder_layer_1_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240696576)))]; tensor linear_15_cast_fp16 = linear(bias = deberta_encoder_layer_1_output_dense_bias_to_fp16, weight = deberta_encoder_layer_1_output_dense_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_15_cast_fp16")]; tensor input_51_cast_fp16 = add(x = linear_15_cast_fp16, y = input_43_cast_fp16)[name = string("input_51_cast_fp16")]; tensor input_53_axes_0 = const()[name = string("input_53_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_1_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_1_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240698176)))]; tensor deberta_encoder_layer_1_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_1_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240699776)))]; tensor input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = deberta_encoder_layer_1_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_1_output_LayerNorm_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; tensor deberta_encoder_layer_2_attention_self_query_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_query_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240701376)))]; tensor deberta_encoder_layer_2_attention_self_query_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_query_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241881088)))]; tensor linear_16_cast_fp16 = linear(bias = deberta_encoder_layer_2_attention_self_query_proj_bias_to_fp16, weight = deberta_encoder_layer_2_attention_self_query_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_16_cast_fp16")]; tensor var_596 = const()[name = string("op_596"), val = tensor([1, 512, 12, -1])]; tensor x_43_cast_fp16 = reshape(shape = var_596, x = linear_16_cast_fp16)[name = string("x_43_cast_fp16")]; tensor var_598 = const()[name = string("op_598"), val = tensor([0, 2, 1, 3])]; tensor var_603 = const()[name = string("op_603"), val = tensor([-1, 512, 64])]; tensor var_599_cast_fp16 = transpose(perm = var_598, x = x_43_cast_fp16)[name = string("transpose_67")]; tensor query_layer_15_cast_fp16 = reshape(shape = var_603, x = var_599_cast_fp16)[name = string("query_layer_15_cast_fp16")]; tensor deberta_encoder_layer_2_attention_self_key_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_key_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241882688)))]; tensor deberta_encoder_layer_2_attention_self_key_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_key_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243062400)))]; tensor linear_17_cast_fp16 = linear(bias = deberta_encoder_layer_2_attention_self_key_proj_bias_to_fp16, weight = deberta_encoder_layer_2_attention_self_key_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_17_cast_fp16")]; tensor var_610 = const()[name = string("op_610"), val = tensor([1, 512, 12, -1])]; tensor x_47_cast_fp16 = reshape(shape = var_610, x = linear_17_cast_fp16)[name = string("x_47_cast_fp16")]; tensor var_612 = const()[name = string("op_612"), val = tensor([0, 2, 1, 3])]; tensor var_617 = const()[name = string("op_617"), val = tensor([-1, 512, 64])]; tensor var_613_cast_fp16 = transpose(perm = var_612, x = x_47_cast_fp16)[name = string("transpose_66")]; tensor key_layer_5_cast_fp16 = reshape(shape = var_617, x = var_613_cast_fp16)[name = string("key_layer_5_cast_fp16")]; tensor deberta_encoder_layer_2_attention_self_value_proj_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_value_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243064000)))]; tensor deberta_encoder_layer_2_attention_self_value_proj_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_self_value_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244243712)))]; tensor linear_18_cast_fp16 = linear(bias = deberta_encoder_layer_2_attention_self_value_proj_bias_to_fp16, weight = deberta_encoder_layer_2_attention_self_value_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_18_cast_fp16")]; tensor var_624 = const()[name = string("op_624"), val = tensor([1, 512, 12, -1])]; tensor x_51_cast_fp16 = reshape(shape = var_624, x = linear_18_cast_fp16)[name = string("x_51_cast_fp16")]; tensor var_626 = const()[name = string("op_626"), val = tensor([0, 2, 1, 3])]; tensor var_631 = const()[name = string("op_631"), val = tensor([-1, 512, 64])]; tensor var_627_cast_fp16 = transpose(perm = var_626, x = x_51_cast_fp16)[name = string("transpose_65")]; tensor value_layer_5_cast_fp16 = reshape(shape = var_631, x = var_627_cast_fp16)[name = string("value_layer_5_cast_fp16")]; tensor var_641_perm_0 = const()[name = string("op_641_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_643_y_0_to_fp16 = const()[name = string("_inversed_643_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor var_641_cast_fp16 = transpose(perm = var_641_perm_0, x = key_layer_5_cast_fp16)[name = string("transpose_64")]; tensor _inversed_643_cast_fp16 = mul(x = var_641_cast_fp16, y = _inversed_643_y_0_to_fp16)[name = string("_inversed_643_cast_fp16")]; bool attention_scores_13_transpose_x_0 = const()[name = string("attention_scores_13_transpose_x_0"), val = bool(false)]; bool attention_scores_13_transpose_y_0 = const()[name = string("attention_scores_13_transpose_y_0"), val = bool(false)]; tensor attention_scores_13_cast_fp16 = matmul(transpose_x = attention_scores_13_transpose_x_0, transpose_y = attention_scores_13_transpose_y_0, x = query_layer_15_cast_fp16, y = _inversed_643_cast_fp16)[name = string("attention_scores_13_cast_fp16")]; bool c2p_att_9_transpose_x_0 = const()[name = string("c2p_att_9_transpose_x_0"), val = bool(false)]; bool c2p_att_9_transpose_y_0 = const()[name = string("c2p_att_9_transpose_y_0"), val = bool(false)]; tensor var_699_to_fp16 = const()[name = string("op_699_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244245312)))]; tensor c2p_att_9_cast_fp16 = matmul(transpose_x = c2p_att_9_transpose_x_0, transpose_y = c2p_att_9_transpose_y_0, x = query_layer_15_cast_fp16, y = var_699_to_fp16)[name = string("c2p_att_9_cast_fp16")]; bool c2p_att_11_validate_indices_0 = const()[name = string("c2p_att_11_validate_indices_0"), val = bool(false)]; tensor c2p_att_11_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_271_to_uint16, validate_indices = c2p_att_11_validate_indices_0, x = c2p_att_9_cast_fp16)[name = string("c2p_att_11_cast_fp16_cast_int16")]; tensor _inversed_712_y_0_to_fp16 = const()[name = string("_inversed_712_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor _inversed_712_cast_fp16 = mul(x = c2p_att_11_cast_fp16_cast_int16, y = _inversed_712_y_0_to_fp16)[name = string("_inversed_712_cast_fp16")]; bool p2c_att_9_transpose_x_0 = const()[name = string("p2c_att_9_transpose_x_0"), val = bool(false)]; bool p2c_att_9_transpose_y_0 = const()[name = string("p2c_att_9_transpose_y_0"), val = bool(false)]; tensor var_727_to_fp16 = const()[name = string("op_727_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245031808)))]; tensor p2c_att_9_cast_fp16 = matmul(transpose_x = p2c_att_9_transpose_x_0, transpose_y = p2c_att_9_transpose_y_0, x = key_layer_5_cast_fp16, y = var_727_to_fp16)[name = string("p2c_att_9_cast_fp16")]; bool var_735_validate_indices_0 = const()[name = string("op_735_validate_indices_0"), val = bool(false)]; tensor var_735_cast_fp16_cast_int16 = gather_along_axis(axis = var_32, indices = var_296_to_uint16, validate_indices = var_735_validate_indices_0, x = p2c_att_9_cast_fp16)[name = string("op_735_cast_fp16_cast_int16")]; tensor p2c_att_11_perm_0 = const()[name = string("p2c_att_11_perm_0"), val = tensor([0, -1, -2])]; tensor _inversed_738_y_0_to_fp16 = const()[name = string("_inversed_738_y_0_to_fp16"), val = tensor([0x1.278p-4])]; tensor p2c_att_11_cast_fp16 = transpose(perm = p2c_att_11_perm_0, x = var_735_cast_fp16_cast_int16)[name = string("transpose_63")]; tensor _inversed_738_cast_fp16 = mul(x = p2c_att_11_cast_fp16, y = _inversed_738_y_0_to_fp16)[name = string("_inversed_738_cast_fp16")]; tensor rel_att_5_cast_fp16 = add(x = _inversed_712_cast_fp16, y = _inversed_738_cast_fp16)[name = string("rel_att_5_cast_fp16")]; tensor attention_scores_15_cast_fp16 = add(x = attention_scores_13_cast_fp16, y = rel_att_5_cast_fp16)[name = string("attention_scores_15_cast_fp16")]; tensor var_743 = const()[name = string("op_743"), val = tensor([-1, 12, 512, 512])]; tensor attention_scores_17_cast_fp16 = reshape(shape = var_743, x = attention_scores_15_cast_fp16)[name = string("attention_scores_17_cast_fp16")]; tensor input_57_cast_fp16 = select(a = var_41_to_fp16, b = attention_scores_17_cast_fp16, cond = var_308)[name = string("input_57_cast_fp16")]; tensor input_59_cast_fp16 = softmax(axis = var_32, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")]; tensor var_752 = const()[name = string("op_752"), val = tensor([-1, 512, 512])]; tensor var_753_cast_fp16 = reshape(shape = var_752, x = input_59_cast_fp16)[name = string("op_753_cast_fp16")]; bool context_layer_9_transpose_x_0 = const()[name = string("context_layer_9_transpose_x_0"), val = bool(false)]; bool context_layer_9_transpose_y_0 = const()[name = string("context_layer_9_transpose_y_0"), val = bool(false)]; tensor context_layer_9_cast_fp16 = matmul(transpose_x = context_layer_9_transpose_x_0, transpose_y = context_layer_9_transpose_y_0, x = var_753_cast_fp16, y = value_layer_5_cast_fp16)[name = string("context_layer_9_cast_fp16")]; tensor var_757 = const()[name = string("op_757"), val = tensor([-1, 12, 512, 64])]; tensor var_758_cast_fp16 = reshape(shape = var_757, x = context_layer_9_cast_fp16)[name = string("op_758_cast_fp16")]; tensor var_759 = const()[name = string("op_759"), val = tensor([0, 2, 1, 3])]; tensor var_764 = const()[name = string("op_764"), val = tensor([1, 512, -1])]; tensor var_760_cast_fp16 = transpose(perm = var_759, x = var_758_cast_fp16)[name = string("transpose_62")]; tensor input_61_cast_fp16 = reshape(shape = var_764, x = var_760_cast_fp16)[name = string("input_61_cast_fp16")]; tensor deberta_encoder_layer_2_attention_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245818304)))]; tensor deberta_encoder_layer_2_attention_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246998016)))]; tensor linear_21_cast_fp16 = linear(bias = deberta_encoder_layer_2_attention_output_dense_bias_to_fp16, weight = deberta_encoder_layer_2_attention_output_dense_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_21_cast_fp16")]; tensor input_65_cast_fp16 = add(x = linear_21_cast_fp16, y = input_53_cast_fp16)[name = string("input_65_cast_fp16")]; tensor input_67_axes_0 = const()[name = string("input_67_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246999616)))]; tensor deberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247001216)))]; tensor input_67_cast_fp16 = layer_norm(axes = input_67_axes_0, beta = deberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; tensor deberta_encoder_layer_2_intermediate_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247002816)))]; tensor deberta_encoder_layer_2_intermediate_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251721472)))]; tensor linear_22_cast_fp16 = linear(bias = deberta_encoder_layer_2_intermediate_dense_bias_to_fp16, weight = deberta_encoder_layer_2_intermediate_dense_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_22_cast_fp16")]; string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = linear_22_cast_fp16)[name = string("input_71_cast_fp16")]; tensor deberta_encoder_layer_2_output_dense_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251727680)))]; tensor deberta_encoder_layer_2_output_dense_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256446336)))]; tensor linear_23_cast_fp16 = linear(bias = deberta_encoder_layer_2_output_dense_bias_to_fp16, weight = deberta_encoder_layer_2_output_dense_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_23_cast_fp16")]; tensor input_75_cast_fp16 = add(x = linear_23_cast_fp16, y = input_67_cast_fp16)[name = string("input_75_cast_fp16")]; tensor input_77_axes_0 = const()[name = string("input_77_axes_0"), val = tensor([-1])]; tensor deberta_encoder_layer_2_output_LayerNorm_weight_to_fp16 = const()[name = string("deberta_encoder_layer_2_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256447936)))]; tensor deberta_encoder_layer_2_output_LayerNorm_bias_to_fp16 = const()[name = string("deberta_encoder_layer_2_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256449536)))]; tensor input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = deberta_encoder_layer_2_output_LayerNorm_bias_to_fp16, epsilon = var_51_to_fp16, gamma = deberta_encoder_layer_2_output_LayerNorm_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; tensor