program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3605.1.1"}, {"coremlc-version", "3600.25.1"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] { func main(tensor attention_mask, tensor input_ids) { int32 var_5 = const()[name = string("op_5"), val = int32(1)]; tensor var_42 = not_equal(x = input_ids, y = var_5)[name = string("op_42")]; string mask_dtype_0 = const()[name = string("mask_dtype_0"), val = string("int32")]; bool var_44_exclusive_0 = const()[name = string("op_44_exclusive_0"), val = bool(false)]; bool var_44_reverse_0 = const()[name = string("op_44_reverse_0"), val = bool(false)]; tensor mask = cast(dtype = mask_dtype_0, x = var_42)[name = string("cast_55")]; tensor var_44 = cumsum(axis = var_5, exclusive = var_44_exclusive_0, reverse = var_44_reverse_0, x = mask)[name = string("op_44")]; tensor incremental_indices = mul(x = var_44, y = mask)[name = string("incremental_indices")]; int32 var_50 = const()[name = string("op_50"), val = int32(1)]; tensor input_3 = add(x = incremental_indices, y = var_50)[name = string("input_3")]; 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 m_roberta_embeddings_word_embeddings_weight_to_fp16 = const()[name = string("m_roberta_embeddings_word_embeddings_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)]; tensor greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = string("greater_equal_0")]; int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(40000)]; tensor add_0 = add(x = input_ids, y = slice_by_index_0)[name = string("add_0")]; tensor select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = string("select_0")]; int32 inputs_embeds_cast_fp16_axis_0 = const()[name = string("inputs_embeds_cast_fp16_axis_0"), val = int32(0)]; tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_cast_fp16_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = select_0, validate_indices = inputs_embeds_validate_indices_0, x = m_roberta_embeddings_word_embeddings_weight_to_fp16)[name = string("inputs_embeds_cast_fp16")]; tensor token_type_embeddings_1_to_fp16 = const()[name = string("token_type_embeddings_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61440128)))]; tensor embeddings_1_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = token_type_embeddings_1_to_fp16)[name = string("embeddings_1_cast_fp16")]; int32 position_embeddings_1_axis_0 = const()[name = string("position_embeddings_1_axis_0"), val = int32(0)]; int32 position_embeddings_1_batch_dims_0 = const()[name = string("position_embeddings_1_batch_dims_0"), val = int32(0)]; bool position_embeddings_1_validate_indices_0 = const()[name = string("position_embeddings_1_validate_indices_0"), val = bool(false)]; tensor m_roberta_embeddings_position_embeddings_weight_to_fp16 = const()[name = string("m_roberta_embeddings_position_embeddings_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62226624)))]; string input_3_to_uint16_dtype_0 = const()[name = string("input_3_to_uint16_dtype_0"), val = string("uint16")]; tensor input_3_to_uint16 = cast(dtype = input_3_to_uint16_dtype_0, x = input_3)[name = string("cast_54")]; tensor position_embeddings_1_cast_fp16_cast_uint16 = gather(axis = position_embeddings_1_axis_0, batch_dims = position_embeddings_1_batch_dims_0, indices = input_3_to_uint16, validate_indices = position_embeddings_1_validate_indices_0, x = m_roberta_embeddings_position_embeddings_weight_to_fp16)[name = string("position_embeddings_1_cast_fp16_cast_uint16")]; tensor input_5_cast_fp16 = add(x = embeddings_1_cast_fp16, y = position_embeddings_1_cast_fp16_cast_uint16)[name = string("input_5_cast_fp16")]; tensor input_7_axes_0 = const()[name = string("input_7_axes_0"), val = tensor([-1])]; tensor m_roberta_embeddings_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_embeddings_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63016192)))]; tensor m_roberta_embeddings_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_embeddings_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63017792)))]; fp16 var_10_to_fp16 = const()[name = string("op_10_to_fp16"), val = fp16(0x1.5p-17)]; tensor input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = m_roberta_embeddings_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_embeddings_LayerNorm_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; tensor var_68_axes_0 = const()[name = string("op_68_axes_0"), val = tensor([1])]; tensor var_68 = expand_dims(axes = var_68_axes_0, x = attention_mask)[name = string("op_68")]; tensor var_69_axes_0 = const()[name = string("op_69_axes_0"), val = tensor([2])]; tensor var_69 = expand_dims(axes = var_69_axes_0, x = var_68)[name = string("op_69")]; tensor var_72_reps_0 = const()[name = string("op_72_reps_0"), val = tensor([1, 1, 512, 1])]; tensor var_72 = tile(reps = var_72_reps_0, x = var_69)[name = string("op_72")]; fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x1p+0)]; string expanded_mask_to_fp16_dtype_0 = const()[name = string("expanded_mask_to_fp16_dtype_0"), val = string("fp16")]; tensor var_72_to_fp16 = cast(dtype = expanded_mask_to_fp16_dtype_0, x = var_72)[name = string("cast_53")]; tensor inverted_mask_cast_fp16 = sub(x = const_4_to_fp16, y = var_72_to_fp16)[name = string("inverted_mask_cast_fp16")]; string var_77_dtype_0 = const()[name = string("op_77_dtype_0"), val = string("bool")]; fp16 var_21_to_fp16 = const()[name = string("op_21_to_fp16"), val = fp16(-inf)]; tensor inverted_mask_cast_fp16_to_bool = cast(dtype = var_77_dtype_0, x = inverted_mask_cast_fp16)[name = string("cast_52")]; tensor attention_mask_cast_fp16 = select(a = var_21_to_fp16, b = inverted_mask_cast_fp16, cond = inverted_mask_cast_fp16_to_bool)[name = string("attention_mask_cast_fp16")]; tensor m_roberta_encoder_layer_0_attention_self_query_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_query_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63019392)))]; tensor m_roberta_encoder_layer_0_attention_self_query_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64199104)))]; tensor linear_0_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_attention_self_query_bias_to_fp16, weight = m_roberta_encoder_layer_0_attention_self_query_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_0_cast_fp16")]; tensor var_116 = const()[name = string("op_116"), val = tensor([1, -1, 12, 64])]; tensor var_117_cast_fp16 = reshape(shape = var_116, x = linear_0_cast_fp16)[name = string("op_117_cast_fp16")]; tensor m_roberta_encoder_layer_0_attention_self_key_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_key_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64200704)))]; tensor m_roberta_encoder_layer_0_attention_self_key_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65380416)))]; tensor linear_1_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_attention_self_key_bias_to_fp16, weight = m_roberta_encoder_layer_0_attention_self_key_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_1_cast_fp16")]; tensor var_122 = const()[name = string("op_122"), val = tensor([1, -1, 12, 64])]; tensor var_123_cast_fp16 = reshape(shape = var_122, x = linear_1_cast_fp16)[name = string("op_123_cast_fp16")]; tensor m_roberta_encoder_layer_0_attention_self_value_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_value_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65382016)))]; tensor m_roberta_encoder_layer_0_attention_self_value_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66561728)))]; tensor linear_2_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_attention_self_value_bias_to_fp16, weight = m_roberta_encoder_layer_0_attention_self_value_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_2_cast_fp16")]; tensor var_128 = const()[name = string("op_128"), val = tensor([1, -1, 12, 64])]; tensor var_129_cast_fp16 = reshape(shape = var_128, x = linear_2_cast_fp16)[name = string("op_129_cast_fp16")]; tensor transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_74 = transpose(perm = transpose_74_perm_0, x = var_129_cast_fp16)[name = string("transpose_153")]; tensor transpose_73 = transpose(perm = transpose_73_perm_0, x = var_123_cast_fp16)[name = string("transpose_154")]; tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = var_117_cast_fp16)[name = string("transpose_155")]; tensor attn_output_1_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_73, query = transpose_72, value = transpose_74)[name = string("attn_output_1_cast_fp16")]; tensor attn_output_3_perm_0 = const()[name = string("attn_output_3_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_133 = const()[name = string("op_133"), val = tensor([1, 512, 768])]; tensor attn_output_3_cast_fp16 = transpose(perm = attn_output_3_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_152")]; tensor input_9_cast_fp16 = reshape(shape = var_133, x = attn_output_3_cast_fp16)[name = string("input_9_cast_fp16")]; tensor m_roberta_encoder_layer_0_attention_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66563328)))]; tensor m_roberta_encoder_layer_0_attention_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67743040)))]; tensor linear_3_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_attention_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_0_attention_output_dense_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")]; tensor input_13_cast_fp16 = add(x = linear_3_cast_fp16, y = input_7_cast_fp16)[name = string("input_13_cast_fp16")]; tensor input_15_axes_0 = const()[name = string("input_15_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67744640)))]; tensor m_roberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67746240)))]; tensor input_15_cast_fp16 = layer_norm(axes = input_15_axes_0, beta = m_roberta_encoder_layer_0_attention_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_0_attention_output_LayerNorm_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; tensor m_roberta_encoder_layer_0_intermediate_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67747840)))]; tensor m_roberta_encoder_layer_0_intermediate_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72466496)))]; tensor linear_4_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_intermediate_dense_bias_to_fp16, weight = m_roberta_encoder_layer_0_intermediate_dense_weight_to_fp16, x = input_15_cast_fp16)[name = string("linear_4_cast_fp16")]; string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_4_cast_fp16)[name = string("input_19_cast_fp16")]; tensor m_roberta_encoder_layer_0_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72472704)))]; tensor m_roberta_encoder_layer_0_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77191360)))]; tensor linear_5_cast_fp16 = linear(bias = m_roberta_encoder_layer_0_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_0_output_dense_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_5_cast_fp16")]; tensor input_23_cast_fp16 = add(x = linear_5_cast_fp16, y = input_15_cast_fp16)[name = string("input_23_cast_fp16")]; tensor hidden_states_7_axes_0 = const()[name = string("hidden_states_7_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_0_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77192960)))]; tensor m_roberta_encoder_layer_0_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_0_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77194560)))]; tensor hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = m_roberta_encoder_layer_0_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_0_output_LayerNorm_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; tensor m_roberta_encoder_layer_1_attention_self_query_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_query_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77196160)))]; tensor m_roberta_encoder_layer_1_attention_self_query_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78375872)))]; tensor linear_6_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_attention_self_query_bias_to_fp16, weight = m_roberta_encoder_layer_1_attention_self_query_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = string("linear_6_cast_fp16")]; tensor var_175 = const()[name = string("op_175"), val = tensor([1, -1, 12, 64])]; tensor var_176_cast_fp16 = reshape(shape = var_175, x = linear_6_cast_fp16)[name = string("op_176_cast_fp16")]; tensor m_roberta_encoder_layer_1_attention_self_key_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_key_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78377472)))]; tensor m_roberta_encoder_layer_1_attention_self_key_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79557184)))]; tensor linear_7_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_attention_self_key_bias_to_fp16, weight = m_roberta_encoder_layer_1_attention_self_key_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = string("linear_7_cast_fp16")]; tensor var_181 = const()[name = string("op_181"), val = tensor([1, -1, 12, 64])]; tensor var_182_cast_fp16 = reshape(shape = var_181, x = linear_7_cast_fp16)[name = string("op_182_cast_fp16")]; tensor m_roberta_encoder_layer_1_attention_self_value_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_value_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79558784)))]; tensor m_roberta_encoder_layer_1_attention_self_value_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80738496)))]; tensor linear_8_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_attention_self_value_bias_to_fp16, weight = m_roberta_encoder_layer_1_attention_self_value_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = string("linear_8_cast_fp16")]; tensor var_187 = const()[name = string("op_187"), val = tensor([1, -1, 12, 64])]; tensor var_188_cast_fp16 = reshape(shape = var_187, x = linear_8_cast_fp16)[name = string("op_188_cast_fp16")]; tensor transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_77 = transpose(perm = transpose_77_perm_0, x = var_188_cast_fp16)[name = string("transpose_149")]; tensor transpose_76 = transpose(perm = transpose_76_perm_0, x = var_182_cast_fp16)[name = string("transpose_150")]; tensor transpose_75 = transpose(perm = transpose_75_perm_0, x = var_176_cast_fp16)[name = string("transpose_151")]; tensor attn_output_5_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_76, query = transpose_75, value = transpose_77)[name = string("attn_output_5_cast_fp16")]; tensor attn_output_7_perm_0 = const()[name = string("attn_output_7_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_192 = const()[name = string("op_192"), val = tensor([1, 512, 768])]; tensor attn_output_7_cast_fp16 = transpose(perm = attn_output_7_perm_0, x = attn_output_5_cast_fp16)[name = string("transpose_148")]; tensor input_25_cast_fp16 = reshape(shape = var_192, x = attn_output_7_cast_fp16)[name = string("input_25_cast_fp16")]; tensor m_roberta_encoder_layer_1_attention_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80740096)))]; tensor m_roberta_encoder_layer_1_attention_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81919808)))]; tensor linear_9_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_attention_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_1_attention_output_dense_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_9_cast_fp16")]; tensor input_29_cast_fp16 = add(x = linear_9_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("input_29_cast_fp16")]; tensor input_31_axes_0 = const()[name = string("input_31_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81921408)))]; tensor m_roberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81923008)))]; tensor input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = m_roberta_encoder_layer_1_attention_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_1_attention_output_LayerNorm_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; tensor m_roberta_encoder_layer_1_intermediate_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81924608)))]; tensor m_roberta_encoder_layer_1_intermediate_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86643264)))]; tensor linear_10_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_intermediate_dense_bias_to_fp16, weight = m_roberta_encoder_layer_1_intermediate_dense_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_10_cast_fp16")]; string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("EXACT")]; tensor input_35_cast_fp16 = gelu(mode = input_35_mode_0, x = linear_10_cast_fp16)[name = string("input_35_cast_fp16")]; tensor m_roberta_encoder_layer_1_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86649472)))]; tensor m_roberta_encoder_layer_1_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91368128)))]; tensor linear_11_cast_fp16 = linear(bias = m_roberta_encoder_layer_1_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_1_output_dense_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_11_cast_fp16")]; tensor input_39_cast_fp16 = add(x = linear_11_cast_fp16, y = input_31_cast_fp16)[name = string("input_39_cast_fp16")]; tensor hidden_states_13_axes_0 = const()[name = string("hidden_states_13_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_1_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91369728)))]; tensor m_roberta_encoder_layer_1_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_1_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91371328)))]; tensor hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = m_roberta_encoder_layer_1_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_1_output_LayerNorm_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; tensor m_roberta_encoder_layer_2_attention_self_query_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_query_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91372928)))]; tensor m_roberta_encoder_layer_2_attention_self_query_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92552640)))]; tensor linear_12_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_attention_self_query_bias_to_fp16, weight = m_roberta_encoder_layer_2_attention_self_query_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = string("linear_12_cast_fp16")]; tensor var_234 = const()[name = string("op_234"), val = tensor([1, -1, 12, 64])]; tensor var_235_cast_fp16 = reshape(shape = var_234, x = linear_12_cast_fp16)[name = string("op_235_cast_fp16")]; tensor m_roberta_encoder_layer_2_attention_self_key_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_key_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92554240)))]; tensor m_roberta_encoder_layer_2_attention_self_key_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93733952)))]; tensor linear_13_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_attention_self_key_bias_to_fp16, weight = m_roberta_encoder_layer_2_attention_self_key_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = string("linear_13_cast_fp16")]; tensor var_240 = const()[name = string("op_240"), val = tensor([1, -1, 12, 64])]; tensor var_241_cast_fp16 = reshape(shape = var_240, x = linear_13_cast_fp16)[name = string("op_241_cast_fp16")]; tensor m_roberta_encoder_layer_2_attention_self_value_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_value_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93735552)))]; tensor m_roberta_encoder_layer_2_attention_self_value_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94915264)))]; tensor linear_14_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_attention_self_value_bias_to_fp16, weight = m_roberta_encoder_layer_2_attention_self_value_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = string("linear_14_cast_fp16")]; tensor var_246 = const()[name = string("op_246"), val = tensor([1, -1, 12, 64])]; tensor var_247_cast_fp16 = reshape(shape = var_246, x = linear_14_cast_fp16)[name = string("op_247_cast_fp16")]; tensor transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_80 = transpose(perm = transpose_80_perm_0, x = var_247_cast_fp16)[name = string("transpose_145")]; tensor transpose_79 = transpose(perm = transpose_79_perm_0, x = var_241_cast_fp16)[name = string("transpose_146")]; tensor transpose_78 = transpose(perm = transpose_78_perm_0, x = var_235_cast_fp16)[name = string("transpose_147")]; tensor attn_output_9_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_79, query = transpose_78, value = transpose_80)[name = string("attn_output_9_cast_fp16")]; tensor attn_output_11_perm_0 = const()[name = string("attn_output_11_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_251 = const()[name = string("op_251"), val = tensor([1, 512, 768])]; tensor attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = string("transpose_144")]; tensor input_41_cast_fp16 = reshape(shape = var_251, x = attn_output_11_cast_fp16)[name = string("input_41_cast_fp16")]; tensor m_roberta_encoder_layer_2_attention_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94916864)))]; tensor m_roberta_encoder_layer_2_attention_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96096576)))]; tensor linear_15_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_attention_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_2_attention_output_dense_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_15_cast_fp16")]; tensor input_45_cast_fp16 = add(x = linear_15_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("input_45_cast_fp16")]; tensor input_47_axes_0 = const()[name = string("input_47_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96098176)))]; tensor m_roberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96099776)))]; tensor input_47_cast_fp16 = layer_norm(axes = input_47_axes_0, beta = m_roberta_encoder_layer_2_attention_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_2_attention_output_LayerNorm_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; tensor m_roberta_encoder_layer_2_intermediate_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96101376)))]; tensor m_roberta_encoder_layer_2_intermediate_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100820032)))]; tensor linear_16_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_intermediate_dense_bias_to_fp16, weight = m_roberta_encoder_layer_2_intermediate_dense_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_16_cast_fp16")]; string input_51_mode_0 = const()[name = string("input_51_mode_0"), val = string("EXACT")]; tensor input_51_cast_fp16 = gelu(mode = input_51_mode_0, x = linear_16_cast_fp16)[name = string("input_51_cast_fp16")]; tensor m_roberta_encoder_layer_2_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100826240)))]; tensor m_roberta_encoder_layer_2_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105544896)))]; tensor linear_17_cast_fp16 = linear(bias = m_roberta_encoder_layer_2_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_2_output_dense_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_17_cast_fp16")]; tensor input_55_cast_fp16 = add(x = linear_17_cast_fp16, y = input_47_cast_fp16)[name = string("input_55_cast_fp16")]; tensor hidden_states_19_axes_0 = const()[name = string("hidden_states_19_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_2_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105546496)))]; tensor m_roberta_encoder_layer_2_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_2_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105548096)))]; tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = m_roberta_encoder_layer_2_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_2_output_LayerNorm_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; tensor m_roberta_encoder_layer_3_attention_self_query_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_query_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105549696)))]; tensor m_roberta_encoder_layer_3_attention_self_query_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106729408)))]; tensor linear_18_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_attention_self_query_bias_to_fp16, weight = m_roberta_encoder_layer_3_attention_self_query_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = string("linear_18_cast_fp16")]; tensor var_293 = const()[name = string("op_293"), val = tensor([1, -1, 12, 64])]; tensor var_294_cast_fp16 = reshape(shape = var_293, x = linear_18_cast_fp16)[name = string("op_294_cast_fp16")]; tensor m_roberta_encoder_layer_3_attention_self_key_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_key_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106731008)))]; tensor m_roberta_encoder_layer_3_attention_self_key_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107910720)))]; tensor linear_19_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_attention_self_key_bias_to_fp16, weight = m_roberta_encoder_layer_3_attention_self_key_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = string("linear_19_cast_fp16")]; tensor var_299 = const()[name = string("op_299"), val = tensor([1, -1, 12, 64])]; tensor var_300_cast_fp16 = reshape(shape = var_299, x = linear_19_cast_fp16)[name = string("op_300_cast_fp16")]; tensor m_roberta_encoder_layer_3_attention_self_value_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_value_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107912320)))]; tensor m_roberta_encoder_layer_3_attention_self_value_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109092032)))]; tensor linear_20_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_attention_self_value_bias_to_fp16, weight = m_roberta_encoder_layer_3_attention_self_value_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = string("linear_20_cast_fp16")]; tensor var_305 = const()[name = string("op_305"), val = tensor([1, -1, 12, 64])]; tensor var_306_cast_fp16 = reshape(shape = var_305, x = linear_20_cast_fp16)[name = string("op_306_cast_fp16")]; tensor transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_83 = transpose(perm = transpose_83_perm_0, x = var_306_cast_fp16)[name = string("transpose_141")]; tensor transpose_82 = transpose(perm = transpose_82_perm_0, x = var_300_cast_fp16)[name = string("transpose_142")]; tensor transpose_81 = transpose(perm = transpose_81_perm_0, x = var_294_cast_fp16)[name = string("transpose_143")]; tensor attn_output_13_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_82, query = transpose_81, value = transpose_83)[name = string("attn_output_13_cast_fp16")]; tensor attn_output_15_perm_0 = const()[name = string("attn_output_15_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_310 = const()[name = string("op_310"), val = tensor([1, 512, 768])]; tensor attn_output_15_cast_fp16 = transpose(perm = attn_output_15_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_140")]; tensor input_57_cast_fp16 = reshape(shape = var_310, x = attn_output_15_cast_fp16)[name = string("input_57_cast_fp16")]; tensor m_roberta_encoder_layer_3_attention_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109093632)))]; tensor m_roberta_encoder_layer_3_attention_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110273344)))]; tensor linear_21_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_attention_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_3_attention_output_dense_weight_to_fp16, x = input_57_cast_fp16)[name = string("linear_21_cast_fp16")]; tensor input_61_cast_fp16 = add(x = linear_21_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("input_61_cast_fp16")]; tensor input_63_axes_0 = const()[name = string("input_63_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110274944)))]; tensor m_roberta_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110276544)))]; tensor input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = m_roberta_encoder_layer_3_attention_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_3_attention_output_LayerNorm_weight_to_fp16, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; tensor m_roberta_encoder_layer_3_intermediate_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110278144)))]; tensor m_roberta_encoder_layer_3_intermediate_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114996800)))]; tensor linear_22_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_intermediate_dense_bias_to_fp16, weight = m_roberta_encoder_layer_3_intermediate_dense_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_22_cast_fp16")]; string input_67_mode_0 = const()[name = string("input_67_mode_0"), val = string("EXACT")]; tensor input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = linear_22_cast_fp16)[name = string("input_67_cast_fp16")]; tensor m_roberta_encoder_layer_3_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115003008)))]; tensor m_roberta_encoder_layer_3_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119721664)))]; tensor linear_23_cast_fp16 = linear(bias = m_roberta_encoder_layer_3_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_3_output_dense_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_23_cast_fp16")]; tensor input_71_cast_fp16 = add(x = linear_23_cast_fp16, y = input_63_cast_fp16)[name = string("input_71_cast_fp16")]; tensor hidden_states_25_axes_0 = const()[name = string("hidden_states_25_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_3_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119723264)))]; tensor m_roberta_encoder_layer_3_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_3_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119724864)))]; tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = m_roberta_encoder_layer_3_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_3_output_LayerNorm_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; tensor m_roberta_encoder_layer_4_attention_self_query_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_query_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119726464)))]; tensor m_roberta_encoder_layer_4_attention_self_query_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_query_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120906176)))]; tensor linear_24_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_attention_self_query_bias_to_fp16, weight = m_roberta_encoder_layer_4_attention_self_query_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = string("linear_24_cast_fp16")]; tensor var_352 = const()[name = string("op_352"), val = tensor([1, -1, 12, 64])]; tensor var_353_cast_fp16 = reshape(shape = var_352, x = linear_24_cast_fp16)[name = string("op_353_cast_fp16")]; tensor m_roberta_encoder_layer_4_attention_self_key_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_key_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120907776)))]; tensor m_roberta_encoder_layer_4_attention_self_key_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_key_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122087488)))]; tensor linear_25_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_attention_self_key_bias_to_fp16, weight = m_roberta_encoder_layer_4_attention_self_key_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = string("linear_25_cast_fp16")]; tensor var_358 = const()[name = string("op_358"), val = tensor([1, -1, 12, 64])]; tensor var_359_cast_fp16 = reshape(shape = var_358, x = linear_25_cast_fp16)[name = string("op_359_cast_fp16")]; tensor m_roberta_encoder_layer_4_attention_self_value_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_value_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122089088)))]; tensor m_roberta_encoder_layer_4_attention_self_value_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_self_value_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123268800)))]; tensor linear_26_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_attention_self_value_bias_to_fp16, weight = m_roberta_encoder_layer_4_attention_self_value_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = string("linear_26_cast_fp16")]; tensor var_364 = const()[name = string("op_364"), val = tensor([1, -1, 12, 64])]; tensor var_365_cast_fp16 = reshape(shape = var_364, x = linear_26_cast_fp16)[name = string("op_365_cast_fp16")]; tensor transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; tensor transpose_86 = transpose(perm = transpose_86_perm_0, x = var_365_cast_fp16)[name = string("transpose_137")]; tensor transpose_85 = transpose(perm = transpose_85_perm_0, x = var_359_cast_fp16)[name = string("transpose_138")]; tensor transpose_84 = transpose(perm = transpose_84_perm_0, x = var_353_cast_fp16)[name = string("transpose_139")]; tensor attn_output_17_cast_fp16 = scaled_dot_product_attention(attn_mask = attention_mask_cast_fp16, key = transpose_85, query = transpose_84, value = transpose_86)[name = string("attn_output_17_cast_fp16")]; tensor attn_output_19_perm_0 = const()[name = string("attn_output_19_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_369 = const()[name = string("op_369"), val = tensor([1, 512, 768])]; tensor attn_output_19_cast_fp16 = transpose(perm = attn_output_19_perm_0, x = attn_output_17_cast_fp16)[name = string("transpose_136")]; tensor input_73_cast_fp16 = reshape(shape = var_369, x = attn_output_19_cast_fp16)[name = string("input_73_cast_fp16")]; tensor m_roberta_encoder_layer_4_attention_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123270400)))]; tensor m_roberta_encoder_layer_4_attention_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124450112)))]; tensor linear_27_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_attention_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_4_attention_output_dense_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_27_cast_fp16")]; tensor input_77_cast_fp16 = add(x = linear_27_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("input_77_cast_fp16")]; tensor input_79_axes_0 = const()[name = string("input_79_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124451712)))]; tensor m_roberta_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124453312)))]; tensor input_79_cast_fp16 = layer_norm(axes = input_79_axes_0, beta = m_roberta_encoder_layer_4_attention_output_LayerNorm_bias_to_fp16, epsilon = var_10_to_fp16, gamma = m_roberta_encoder_layer_4_attention_output_LayerNorm_weight_to_fp16, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; tensor m_roberta_encoder_layer_4_intermediate_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_intermediate_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124454912)))]; tensor m_roberta_encoder_layer_4_intermediate_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_intermediate_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129173568)))]; tensor linear_28_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_intermediate_dense_bias_to_fp16, weight = m_roberta_encoder_layer_4_intermediate_dense_weight_to_fp16, x = input_79_cast_fp16)[name = string("linear_28_cast_fp16")]; string input_83_mode_0 = const()[name = string("input_83_mode_0"), val = string("EXACT")]; tensor input_83_cast_fp16 = gelu(mode = input_83_mode_0, x = linear_28_cast_fp16)[name = string("input_83_cast_fp16")]; tensor m_roberta_encoder_layer_4_output_dense_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_output_dense_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129179776)))]; tensor m_roberta_encoder_layer_4_output_dense_bias_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_output_dense_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133898432)))]; tensor linear_29_cast_fp16 = linear(bias = m_roberta_encoder_layer_4_output_dense_bias_to_fp16, weight = m_roberta_encoder_layer_4_output_dense_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_29_cast_fp16")]; tensor input_87_cast_fp16 = add(x = linear_29_cast_fp16, y = input_79_cast_fp16)[name = string("input_87_cast_fp16")]; tensor hidden_states_31_axes_0 = const()[name = string("hidden_states_31_axes_0"), val = tensor([-1])]; tensor m_roberta_encoder_layer_4_output_LayerNorm_weight_to_fp16 = const()[name = string("m_roberta_encoder_layer_4_output_LayerNorm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133900032)))]; tensor