| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] |
| { |
| func main<ios16>(tensor<fp32, [1, 128, 3000]> mel_spectrogram) { |
| tensor<string, []> var_95_pad_type_0 = const()[name = tensor<string, []>("op_95_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_95_pad_0 = const()[name = tensor<string, []>("op_95_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_95_strides_0 = const()[name = tensor<string, []>("op_95_strides_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1]> var_95_dilations_0 = const()[name = tensor<string, []>("op_95_dilations_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, []> var_95_groups_0 = const()[name = tensor<string, []>("op_95_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<string, []> mel_spectrogram_to_fp16_dtype_0 = const()[name = tensor<string, []>("mel_spectrogram_to_fp16_dtype_0"), val = tensor<string, []>("fp16")]; |
| tensor<fp16, [1280, 128, 3]> encoder_conv1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 128, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<fp16, [1280]> encoder_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(983168)))]; |
| tensor<fp16, [1, 128, 3000]> mel_spectrogram_to_fp16 = cast(dtype = mel_spectrogram_to_fp16_dtype_0, x = mel_spectrogram)[name = tensor<string, []>("cast_132")]; |
| tensor<fp16, [1, 1280, 3000]> var_95_cast_fp16 = conv(bias = encoder_conv1_bias_to_fp16, dilations = var_95_dilations_0, groups = var_95_groups_0, pad = var_95_pad_0, pad_type = var_95_pad_type_0, strides = var_95_strides_0, weight = encoder_conv1_weight_to_fp16, x = mel_spectrogram_to_fp16)[name = tensor<string, []>("op_95_cast_fp16")]; |
| tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1280, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_95_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| tensor<string, []> var_103_pad_type_0 = const()[name = tensor<string, []>("op_103_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_103_pad_0 = const()[name = tensor<string, []>("op_103_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_103_strides_0 = const()[name = tensor<string, []>("op_103_strides_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1]> var_103_dilations_0 = const()[name = tensor<string, []>("op_103_dilations_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, []> var_103_groups_0 = const()[name = tensor<string, []>("op_103_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1280, 1280, 3]> encoder_conv2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985792)))]; |
| tensor<fp16, [1280]> encoder_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10816256)))]; |
| tensor<fp16, [1, 1280, 1500]> var_103_cast_fp16 = conv(bias = encoder_conv2_bias_to_fp16, dilations = var_103_dilations_0, groups = var_103_groups_0, pad = var_103_pad_0, pad_type = var_103_pad_type_0, strides = var_103_strides_0, weight = encoder_conv2_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_103_cast_fp16")]; |
| tensor<string, []> hidden_states_1_mode_0 = const()[name = tensor<string, []>("hidden_states_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1280, 1500]> hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_103_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; |
| tensor<int32, [3]> hidden_states_3_perm_0 = const()[name = tensor<string, []>("hidden_states_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<fp16, [1500, 1280]> encoder_embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("encoder_embed_positions_weight_to_fp16"), val = tensor<fp16, [1500, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10818880)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_3_cast_fp16 = transpose(perm = hidden_states_3_perm_0, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("transpose_320")]; |
| tensor<fp16, [1, 1500, 1280]> input_5_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = encoder_embed_positions_weight_to_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_5_axes_0 = const()[name = tensor<string, []>("hidden_states_5_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_0_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14658944)))]; |
| tensor<fp16, [1280]> encoder_layers_0_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14661568)))]; |
| tensor<fp16, []> var_4_to_fp16 = const()[name = tensor<string, []>("op_4_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_5_cast_fp16 = layer_norm(axes = hidden_states_5_axes_0, beta = encoder_layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_0_self_attn_layer_norm_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14664192)))]; |
| tensor<fp16, [1280]> encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17941056)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_0_cast_fp16 = linear(bias = encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17943680)))]; |
| tensor<fp16, [1280]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21220544)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21223168)))]; |
| tensor<fp16, [1280]> encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24500032)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_2_cast_fp16 = linear(bias = encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; |
| tensor<int32, [4]> var_133 = const()[name = tensor<string, []>("op_133"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_134_cast_fp16 = reshape(shape = var_133, x = linear_0_cast_fp16)[name = tensor<string, []>("op_134_cast_fp16")]; |
| tensor<int32, [4]> var_136 = const()[name = tensor<string, []>("op_136"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_137_cast_fp16 = reshape(shape = var_136, x = linear_1_cast_fp16)[name = tensor<string, []>("op_137_cast_fp16")]; |
| tensor<int32, [4]> var_139 = const()[name = tensor<string, []>("op_139"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_140_cast_fp16 = reshape(shape = var_139, x = linear_2_cast_fp16)[name = tensor<string, []>("op_140_cast_fp16")]; |
| tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, []> var_6_to_fp16 = const()[name = tensor<string, []>("op_6_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_0_cast_fp16 = mul(x = var_134_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_0_cast_fp16")]; |
| tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_128_perm_0 = const()[name = tensor<string, []>("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_129 = transpose(perm = transpose_129_perm_0, x = var_137_cast_fp16)[name = tensor<string, []>("transpose_317")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_128 = transpose(perm = transpose_128_perm_0, x = mul_0_cast_fp16)[name = tensor<string, []>("transpose_318")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_128, y = transpose_129)[name = tensor<string, []>("matmul_0_cast_fp16")]; |
| tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = matmul_0_cast_fp16)[name = tensor<string, []>("softmax_0_cast_fp16")]; |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = var_140_cast_fp16)[name = tensor<string, []>("transpose_319")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = value_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; |
| tensor<int32, [4]> var_143_perm_0 = const()[name = tensor<string, []>("op_143_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_145 = const()[name = tensor<string, []>("op_145"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_143_cast_fp16 = transpose(perm = var_143_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_316")]; |
| tensor<fp16, [1, 1500, 1280]> input_7_cast_fp16 = reshape(shape = var_145, x = var_143_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24502656)))]; |
| tensor<fp16, [1280]> encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27779520)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_3_cast_fp16 = linear(bias = encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_9_cast_fp16 = add(x = input_5_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| tensor<int32, [1]> input_11_axes_0 = const()[name = tensor<string, []>("input_11_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_0_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27782144)))]; |
| tensor<fp16, [1280]> encoder_layers_0_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27784768)))]; |
| tensor<fp16, [1, 1500, 1280]> input_11_cast_fp16 = layer_norm(axes = input_11_axes_0, beta = encoder_layers_0_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_0_final_layer_norm_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27787392)))]; |
| tensor<fp16, [5120]> encoder_layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40894656)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_4_cast_fp16 = linear(bias = encoder_layers_0_fc1_bias_to_fp16, weight = encoder_layers_0_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; |
| tensor<string, []> input_13_mode_0 = const()[name = tensor<string, []>("input_13_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_13_cast_fp16 = gelu(mode = input_13_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40904960)))]; |
| tensor<fp16, [1280]> encoder_layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54012224)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_5_cast_fp16 = linear(bias = encoder_layers_0_fc2_bias_to_fp16, weight = encoder_layers_0_fc2_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_15_cast_fp16 = add(x = input_9_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_11_axes_0 = const()[name = tensor<string, []>("hidden_states_11_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_1_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54014848)))]; |
| tensor<fp16, [1280]> encoder_layers_1_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54017472)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_11_cast_fp16 = layer_norm(axes = hidden_states_11_axes_0, beta = encoder_layers_1_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_1_self_attn_layer_norm_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54020096)))]; |
| tensor<fp16, [1280]> encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57296960)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_6_cast_fp16 = linear(bias = encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57299584)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_1_self_attn_k_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60576448)))]; |
| tensor<fp16, [1280]> encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63853312)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_8_cast_fp16 = linear(bias = encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_v_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; |
| tensor<int32, [4]> var_186 = const()[name = tensor<string, []>("op_186"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_187_cast_fp16 = reshape(shape = var_186, x = linear_6_cast_fp16)[name = tensor<string, []>("op_187_cast_fp16")]; |
| tensor<int32, [4]> var_189 = const()[name = tensor<string, []>("op_189"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_190_cast_fp16 = reshape(shape = var_189, x = linear_7_cast_fp16)[name = tensor<string, []>("op_190_cast_fp16")]; |
| tensor<int32, [4]> var_192 = const()[name = tensor<string, []>("op_192"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_193_cast_fp16 = reshape(shape = var_192, x = linear_8_cast_fp16)[name = tensor<string, []>("op_193_cast_fp16")]; |
| tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_1_cast_fp16 = mul(x = var_187_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_1_cast_fp16")]; |
| tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_130_perm_0 = const()[name = tensor<string, []>("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_131 = transpose(perm = transpose_131_perm_0, x = var_190_cast_fp16)[name = tensor<string, []>("transpose_313")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_130 = transpose(perm = transpose_130_perm_0, x = mul_1_cast_fp16)[name = tensor<string, []>("transpose_314")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_130, y = transpose_131)[name = tensor<string, []>("matmul_1_cast_fp16")]; |
| tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = matmul_1_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")]; |
| tensor<bool, []> attn_output_3_transpose_x_0 = const()[name = tensor<string, []>("attn_output_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_3_transpose_y_0 = const()[name = tensor<string, []>("attn_output_3_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = var_193_cast_fp16)[name = tensor<string, []>("transpose_315")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_3_cast_fp16 = matmul(transpose_x = attn_output_3_transpose_x_0, transpose_y = attn_output_3_transpose_y_0, x = softmax_1_cast_fp16, y = value_7_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; |
| tensor<int32, [4]> var_196_perm_0 = const()[name = tensor<string, []>("op_196_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_198 = const()[name = tensor<string, []>("op_198"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_196_cast_fp16 = transpose(perm = var_196_perm_0, x = attn_output_3_cast_fp16)[name = tensor<string, []>("transpose_312")]; |
| tensor<fp16, [1, 1500, 1280]> input_17_cast_fp16 = reshape(shape = var_198, x = var_196_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63855936)))]; |
| tensor<fp16, [1280]> encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67132800)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_9_cast_fp16 = linear(bias = encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_out_proj_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_1_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67135424)))]; |
| tensor<fp16, [1280]> encoder_layers_1_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67138048)))]; |
| tensor<fp16, [1, 1500, 1280]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = encoder_layers_1_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_1_final_layer_norm_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67140672)))]; |
| tensor<fp16, [5120]> encoder_layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80247936)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_10_cast_fp16 = linear(bias = encoder_layers_1_fc1_bias_to_fp16, weight = encoder_layers_1_fc1_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; |
| tensor<string, []> input_23_mode_0 = const()[name = tensor<string, []>("input_23_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80258240)))]; |
| tensor<fp16, [1280]> encoder_layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93365504)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_11_cast_fp16 = linear(bias = encoder_layers_1_fc2_bias_to_fp16, weight = encoder_layers_1_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_25_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_17_axes_0 = const()[name = tensor<string, []>("hidden_states_17_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_2_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93368128)))]; |
| tensor<fp16, [1280]> encoder_layers_2_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93370752)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_17_cast_fp16 = layer_norm(axes = hidden_states_17_axes_0, beta = encoder_layers_2_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_2_self_attn_layer_norm_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93373376)))]; |
| tensor<fp16, [1280]> encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96650240)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_12_cast_fp16 = linear(bias = encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96652864)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99929728)))]; |
| tensor<fp16, [1280]> encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103206592)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_14_cast_fp16 = linear(bias = encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; |
| tensor<int32, [4]> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_240_cast_fp16 = reshape(shape = var_239, x = linear_12_cast_fp16)[name = tensor<string, []>("op_240_cast_fp16")]; |
| tensor<int32, [4]> var_242 = const()[name = tensor<string, []>("op_242"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_243_cast_fp16 = reshape(shape = var_242, x = linear_13_cast_fp16)[name = tensor<string, []>("op_243_cast_fp16")]; |
| tensor<int32, [4]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_246_cast_fp16 = reshape(shape = var_245, x = linear_14_cast_fp16)[name = tensor<string, []>("op_246_cast_fp16")]; |
| tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_2_cast_fp16 = mul(x = var_240_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_2_cast_fp16")]; |
| tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_132_perm_0 = const()[name = tensor<string, []>("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_133 = transpose(perm = transpose_133_perm_0, x = var_243_cast_fp16)[name = tensor<string, []>("transpose_309")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_132 = transpose(perm = transpose_132_perm_0, x = mul_2_cast_fp16)[name = tensor<string, []>("transpose_310")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_132, y = transpose_133)[name = tensor<string, []>("matmul_2_cast_fp16")]; |
| tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = matmul_2_cast_fp16)[name = tensor<string, []>("softmax_2_cast_fp16")]; |
| tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = var_246_cast_fp16)[name = tensor<string, []>("transpose_311")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_2_cast_fp16, y = value_11_cast_fp16)[name = tensor<string, []>("attn_output_5_cast_fp16")]; |
| tensor<int32, [4]> var_249_perm_0 = const()[name = tensor<string, []>("op_249_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_251 = const()[name = tensor<string, []>("op_251"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_249_cast_fp16 = transpose(perm = var_249_perm_0, x = attn_output_5_cast_fp16)[name = tensor<string, []>("transpose_308")]; |
| tensor<fp16, [1, 1500, 1280]> input_27_cast_fp16 = reshape(shape = var_251, x = var_249_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103209216)))]; |
| tensor<fp16, [1280]> encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106486080)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_15_cast_fp16 = linear(bias = encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_out_proj_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_29_cast_fp16 = add(x = input_25_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| tensor<int32, [1]> input_31_axes_0 = const()[name = tensor<string, []>("input_31_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_2_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106488704)))]; |
| tensor<fp16, [1280]> encoder_layers_2_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106491328)))]; |
| tensor<fp16, [1, 1500, 1280]> input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = encoder_layers_2_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_2_final_layer_norm_weight_to_fp16, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106493952)))]; |
| tensor<fp16, [5120]> encoder_layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119601216)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_16_cast_fp16 = linear(bias = encoder_layers_2_fc1_bias_to_fp16, weight = encoder_layers_2_fc1_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; |
| tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_33_cast_fp16 = gelu(mode = input_33_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119611520)))]; |
| tensor<fp16, [1280]> encoder_layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132718784)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_17_cast_fp16 = linear(bias = encoder_layers_2_fc2_bias_to_fp16, weight = encoder_layers_2_fc2_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_35_cast_fp16 = add(x = input_29_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_23_axes_0 = const()[name = tensor<string, []>("hidden_states_23_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_3_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132721408)))]; |
| tensor<fp16, [1280]> encoder_layers_3_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132724032)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_23_cast_fp16 = layer_norm(axes = hidden_states_23_axes_0, beta = encoder_layers_3_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_3_self_attn_layer_norm_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132726656)))]; |
| tensor<fp16, [1280]> encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136003520)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_18_cast_fp16 = linear(bias = encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136006144)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139283008)))]; |
| tensor<fp16, [1280]> encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142559872)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_20_cast_fp16 = linear(bias = encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; |
| tensor<int32, [4]> var_292 = const()[name = tensor<string, []>("op_292"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_293_cast_fp16 = reshape(shape = var_292, x = linear_18_cast_fp16)[name = tensor<string, []>("op_293_cast_fp16")]; |
| tensor<int32, [4]> var_295 = const()[name = tensor<string, []>("op_295"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_296_cast_fp16 = reshape(shape = var_295, x = linear_19_cast_fp16)[name = tensor<string, []>("op_296_cast_fp16")]; |
| tensor<int32, [4]> var_298 = const()[name = tensor<string, []>("op_298"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_299_cast_fp16 = reshape(shape = var_298, x = linear_20_cast_fp16)[name = tensor<string, []>("op_299_cast_fp16")]; |
| tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_3_cast_fp16 = mul(x = var_293_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_3_cast_fp16")]; |
| tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_135 = transpose(perm = transpose_135_perm_0, x = var_296_cast_fp16)[name = tensor<string, []>("transpose_305")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_134 = transpose(perm = transpose_134_perm_0, x = mul_3_cast_fp16)[name = tensor<string, []>("transpose_306")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_134, y = transpose_135)[name = tensor<string, []>("matmul_3_cast_fp16")]; |
| tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = matmul_3_cast_fp16)[name = tensor<string, []>("softmax_3_cast_fp16")]; |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = var_299_cast_fp16)[name = tensor<string, []>("transpose_307")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = softmax_3_cast_fp16, y = value_15_cast_fp16)[name = tensor<string, []>("attn_output_7_cast_fp16")]; |
| tensor<int32, [4]> var_302_perm_0 = const()[name = tensor<string, []>("op_302_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_304 = const()[name = tensor<string, []>("op_304"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_302_cast_fp16 = transpose(perm = var_302_perm_0, x = attn_output_7_cast_fp16)[name = tensor<string, []>("transpose_304")]; |
| tensor<fp16, [1, 1500, 1280]> input_37_cast_fp16 = reshape(shape = var_304, x = var_302_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142562496)))]; |
| tensor<fp16, [1280]> encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145839360)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_21_cast_fp16 = linear(bias = encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_out_proj_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_39_cast_fp16 = add(x = input_35_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| tensor<int32, [1]> input_41_axes_0 = const()[name = tensor<string, []>("input_41_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_3_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145841984)))]; |
| tensor<fp16, [1280]> encoder_layers_3_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145844608)))]; |
| tensor<fp16, [1, 1500, 1280]> input_41_cast_fp16 = layer_norm(axes = input_41_axes_0, beta = encoder_layers_3_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_3_final_layer_norm_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145847232)))]; |
| tensor<fp16, [5120]> encoder_layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158954496)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_22_cast_fp16 = linear(bias = encoder_layers_3_fc1_bias_to_fp16, weight = encoder_layers_3_fc1_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; |
| tensor<string, []> input_43_mode_0 = const()[name = tensor<string, []>("input_43_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_43_cast_fp16 = gelu(mode = input_43_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158964800)))]; |
| tensor<fp16, [1280]> encoder_layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172072064)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_23_cast_fp16 = linear(bias = encoder_layers_3_fc2_bias_to_fp16, weight = encoder_layers_3_fc2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_45_cast_fp16 = add(x = input_39_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_29_axes_0 = const()[name = tensor<string, []>("hidden_states_29_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_4_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172074688)))]; |
| tensor<fp16, [1280]> encoder_layers_4_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172077312)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_29_cast_fp16 = layer_norm(axes = hidden_states_29_axes_0, beta = encoder_layers_4_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_4_self_attn_layer_norm_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("hidden_states_29_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172079936)))]; |
| tensor<fp16, [1280]> encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175356800)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_24_cast_fp16 = linear(bias = encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175359424)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178636288)))]; |
| tensor<fp16, [1280]> encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181913152)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_26_cast_fp16 = linear(bias = encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_states_29_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; |
| tensor<int32, [4]> var_345 = const()[name = tensor<string, []>("op_345"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_346_cast_fp16 = reshape(shape = var_345, x = linear_24_cast_fp16)[name = tensor<string, []>("op_346_cast_fp16")]; |
| tensor<int32, [4]> var_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_349_cast_fp16 = reshape(shape = var_348, x = linear_25_cast_fp16)[name = tensor<string, []>("op_349_cast_fp16")]; |
| tensor<int32, [4]> var_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_352_cast_fp16 = reshape(shape = var_351, x = linear_26_cast_fp16)[name = tensor<string, []>("op_352_cast_fp16")]; |
| tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_4_cast_fp16 = mul(x = var_346_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_4_cast_fp16")]; |
| tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_136_perm_0 = const()[name = tensor<string, []>("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_137 = transpose(perm = transpose_137_perm_0, x = var_349_cast_fp16)[name = tensor<string, []>("transpose_301")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_136 = transpose(perm = transpose_136_perm_0, x = mul_4_cast_fp16)[name = tensor<string, []>("transpose_302")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_136, y = transpose_137)[name = tensor<string, []>("matmul_4_cast_fp16")]; |
| tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = matmul_4_cast_fp16)[name = tensor<string, []>("softmax_4_cast_fp16")]; |
| tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = var_352_cast_fp16)[name = tensor<string, []>("transpose_303")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_4_cast_fp16, y = value_19_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; |
| tensor<int32, [4]> var_355_perm_0 = const()[name = tensor<string, []>("op_355_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_357 = const()[name = tensor<string, []>("op_357"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_355_cast_fp16 = transpose(perm = var_355_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_300")]; |
| tensor<fp16, [1, 1500, 1280]> input_47_cast_fp16 = reshape(shape = var_357, x = var_355_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181915776)))]; |
| tensor<fp16, [1280]> encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185192640)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_27_cast_fp16 = linear(bias = encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_out_proj_weight_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_49_cast_fp16 = add(x = input_45_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| tensor<int32, [1]> input_51_axes_0 = const()[name = tensor<string, []>("input_51_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_4_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185195264)))]; |
| tensor<fp16, [1280]> encoder_layers_4_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185197888)))]; |
| tensor<fp16, [1, 1500, 1280]> input_51_cast_fp16 = layer_norm(axes = input_51_axes_0, beta = encoder_layers_4_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_4_final_layer_norm_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185200512)))]; |
| tensor<fp16, [5120]> encoder_layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198307776)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_28_cast_fp16 = linear(bias = encoder_layers_4_fc1_bias_to_fp16, weight = encoder_layers_4_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; |
| tensor<string, []> input_53_mode_0 = const()[name = tensor<string, []>("input_53_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_53_cast_fp16 = gelu(mode = input_53_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198318080)))]; |
| tensor<fp16, [1280]> encoder_layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211425344)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_29_cast_fp16 = linear(bias = encoder_layers_4_fc2_bias_to_fp16, weight = encoder_layers_4_fc2_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_55_cast_fp16 = add(x = input_49_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_35_axes_0 = const()[name = tensor<string, []>("hidden_states_35_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_5_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211427968)))]; |
| tensor<fp16, [1280]> encoder_layers_5_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211430592)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_35_cast_fp16 = layer_norm(axes = hidden_states_35_axes_0, beta = encoder_layers_5_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_5_self_attn_layer_norm_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("hidden_states_35_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211433216)))]; |
| tensor<fp16, [1280]> encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214710080)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_30_cast_fp16 = linear(bias = encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214712704)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217989568)))]; |
| tensor<fp16, [1280]> encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221266432)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_32_cast_fp16 = linear(bias = encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; |
| tensor<int32, [4]> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_399_cast_fp16 = reshape(shape = var_398, x = linear_30_cast_fp16)[name = tensor<string, []>("op_399_cast_fp16")]; |
| tensor<int32, [4]> var_401 = const()[name = tensor<string, []>("op_401"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_402_cast_fp16 = reshape(shape = var_401, x = linear_31_cast_fp16)[name = tensor<string, []>("op_402_cast_fp16")]; |
| tensor<int32, [4]> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_405_cast_fp16 = reshape(shape = var_404, x = linear_32_cast_fp16)[name = tensor<string, []>("op_405_cast_fp16")]; |
| tensor<int32, [4]> value_23_perm_0 = const()[name = tensor<string, []>("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_5_cast_fp16 = mul(x = var_399_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_5_cast_fp16")]; |
| tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_138_perm_0 = const()[name = tensor<string, []>("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_139_perm_0 = const()[name = tensor<string, []>("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_139 = transpose(perm = transpose_139_perm_0, x = var_402_cast_fp16)[name = tensor<string, []>("transpose_297")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_138 = transpose(perm = transpose_138_perm_0, x = mul_5_cast_fp16)[name = tensor<string, []>("transpose_298")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_138, y = transpose_139)[name = tensor<string, []>("matmul_5_cast_fp16")]; |
| tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = matmul_5_cast_fp16)[name = tensor<string, []>("softmax_5_cast_fp16")]; |
| tensor<bool, []> attn_output_11_transpose_x_0 = const()[name = tensor<string, []>("attn_output_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_11_transpose_y_0 = const()[name = tensor<string, []>("attn_output_11_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = var_405_cast_fp16)[name = tensor<string, []>("transpose_299")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_11_cast_fp16 = matmul(transpose_x = attn_output_11_transpose_x_0, transpose_y = attn_output_11_transpose_y_0, x = softmax_5_cast_fp16, y = value_23_cast_fp16)[name = tensor<string, []>("attn_output_11_cast_fp16")]; |
| tensor<int32, [4]> var_408_perm_0 = const()[name = tensor<string, []>("op_408_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_410 = const()[name = tensor<string, []>("op_410"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_408_cast_fp16 = transpose(perm = var_408_perm_0, x = attn_output_11_cast_fp16)[name = tensor<string, []>("transpose_296")]; |
| tensor<fp16, [1, 1500, 1280]> input_57_cast_fp16 = reshape(shape = var_410, x = var_408_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221269056)))]; |
| tensor<fp16, [1280]> encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224545920)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_33_cast_fp16 = linear(bias = encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_out_proj_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_59_cast_fp16 = add(x = input_55_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_5_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224548544)))]; |
| tensor<fp16, [1280]> encoder_layers_5_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224551168)))]; |
| tensor<fp16, [1, 1500, 1280]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = encoder_layers_5_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_5_final_layer_norm_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224553792)))]; |
| tensor<fp16, [5120]> encoder_layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237661056)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_34_cast_fp16 = linear(bias = encoder_layers_5_fc1_bias_to_fp16, weight = encoder_layers_5_fc1_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; |
| tensor<string, []> input_63_mode_0 = const()[name = tensor<string, []>("input_63_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237671360)))]; |
| tensor<fp16, [1280]> encoder_layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250778624)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_35_cast_fp16 = linear(bias = encoder_layers_5_fc2_bias_to_fp16, weight = encoder_layers_5_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_65_cast_fp16 = add(x = input_59_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_41_axes_0 = const()[name = tensor<string, []>("hidden_states_41_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_6_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250781248)))]; |
| tensor<fp16, [1280]> encoder_layers_6_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250783872)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_41_cast_fp16 = layer_norm(axes = hidden_states_41_axes_0, beta = encoder_layers_6_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_6_self_attn_layer_norm_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("hidden_states_41_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250786496)))]; |
| tensor<fp16, [1280]> encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254063360)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_36_cast_fp16 = linear(bias = encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254065984)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257342848)))]; |
| tensor<fp16, [1280]> encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260619712)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_38_cast_fp16 = linear(bias = encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; |
| tensor<int32, [4]> var_451 = const()[name = tensor<string, []>("op_451"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_452_cast_fp16 = reshape(shape = var_451, x = linear_36_cast_fp16)[name = tensor<string, []>("op_452_cast_fp16")]; |
| tensor<int32, [4]> var_454 = const()[name = tensor<string, []>("op_454"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_455_cast_fp16 = reshape(shape = var_454, x = linear_37_cast_fp16)[name = tensor<string, []>("op_455_cast_fp16")]; |
| tensor<int32, [4]> var_457 = const()[name = tensor<string, []>("op_457"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_458_cast_fp16 = reshape(shape = var_457, x = linear_38_cast_fp16)[name = tensor<string, []>("op_458_cast_fp16")]; |
| tensor<int32, [4]> value_27_perm_0 = const()[name = tensor<string, []>("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_6_cast_fp16 = mul(x = var_452_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_6_cast_fp16")]; |
| tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_140_perm_0 = const()[name = tensor<string, []>("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_141 = transpose(perm = transpose_141_perm_0, x = var_455_cast_fp16)[name = tensor<string, []>("transpose_293")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_140 = transpose(perm = transpose_140_perm_0, x = mul_6_cast_fp16)[name = tensor<string, []>("transpose_294")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_140, y = transpose_141)[name = tensor<string, []>("matmul_6_cast_fp16")]; |
| tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = matmul_6_cast_fp16)[name = tensor<string, []>("softmax_6_cast_fp16")]; |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = var_458_cast_fp16)[name = tensor<string, []>("transpose_295")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_6_cast_fp16, y = value_27_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; |
| tensor<int32, [4]> var_461_perm_0 = const()[name = tensor<string, []>("op_461_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_463 = const()[name = tensor<string, []>("op_463"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_461_cast_fp16 = transpose(perm = var_461_perm_0, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_292")]; |
| tensor<fp16, [1, 1500, 1280]> input_67_cast_fp16 = reshape(shape = var_463, x = var_461_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260622336)))]; |
| tensor<fp16, [1280]> encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263899200)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_39_cast_fp16 = linear(bias = encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_out_proj_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_69_cast_fp16 = add(x = input_65_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| tensor<int32, [1]> input_71_axes_0 = const()[name = tensor<string, []>("input_71_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_6_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263901824)))]; |
| tensor<fp16, [1280]> encoder_layers_6_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263904448)))]; |
| tensor<fp16, [1, 1500, 1280]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = encoder_layers_6_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_6_final_layer_norm_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_6_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263907072)))]; |
| tensor<fp16, [5120]> encoder_layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277014336)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_40_cast_fp16 = linear(bias = encoder_layers_6_fc1_bias_to_fp16, weight = encoder_layers_6_fc1_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; |
| tensor<string, []> input_73_mode_0 = const()[name = tensor<string, []>("input_73_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_73_cast_fp16 = gelu(mode = input_73_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_6_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277024640)))]; |
| tensor<fp16, [1280]> encoder_layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290131904)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_41_cast_fp16 = linear(bias = encoder_layers_6_fc2_bias_to_fp16, weight = encoder_layers_6_fc2_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_75_cast_fp16 = add(x = input_69_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_47_axes_0 = const()[name = tensor<string, []>("hidden_states_47_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_7_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290134528)))]; |
| tensor<fp16, [1280]> encoder_layers_7_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290137152)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = encoder_layers_7_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_7_self_attn_layer_norm_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("hidden_states_47_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290139776)))]; |
| tensor<fp16, [1280]> encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293416640)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_42_cast_fp16 = linear(bias = encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293419264)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296696128)))]; |
| tensor<fp16, [1280]> encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299972992)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_44_cast_fp16 = linear(bias = encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; |
| tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_505_cast_fp16 = reshape(shape = var_504, x = linear_42_cast_fp16)[name = tensor<string, []>("op_505_cast_fp16")]; |
| tensor<int32, [4]> var_507 = const()[name = tensor<string, []>("op_507"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_508_cast_fp16 = reshape(shape = var_507, x = linear_43_cast_fp16)[name = tensor<string, []>("op_508_cast_fp16")]; |
| tensor<int32, [4]> var_510 = const()[name = tensor<string, []>("op_510"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_511_cast_fp16 = reshape(shape = var_510, x = linear_44_cast_fp16)[name = tensor<string, []>("op_511_cast_fp16")]; |
| tensor<int32, [4]> value_31_perm_0 = const()[name = tensor<string, []>("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_7_cast_fp16 = mul(x = var_505_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_7_cast_fp16")]; |
| tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_142_perm_0 = const()[name = tensor<string, []>("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_143_perm_0 = const()[name = tensor<string, []>("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_143 = transpose(perm = transpose_143_perm_0, x = var_508_cast_fp16)[name = tensor<string, []>("transpose_289")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_142 = transpose(perm = transpose_142_perm_0, x = mul_7_cast_fp16)[name = tensor<string, []>("transpose_290")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_142, y = transpose_143)[name = tensor<string, []>("matmul_7_cast_fp16")]; |
| tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = matmul_7_cast_fp16)[name = tensor<string, []>("softmax_7_cast_fp16")]; |
| tensor<bool, []> attn_output_15_transpose_x_0 = const()[name = tensor<string, []>("attn_output_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_15_transpose_y_0 = const()[name = tensor<string, []>("attn_output_15_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = var_511_cast_fp16)[name = tensor<string, []>("transpose_291")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_15_cast_fp16 = matmul(transpose_x = attn_output_15_transpose_x_0, transpose_y = attn_output_15_transpose_y_0, x = softmax_7_cast_fp16, y = value_31_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; |
| tensor<int32, [4]> var_514_perm_0 = const()[name = tensor<string, []>("op_514_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_516 = const()[name = tensor<string, []>("op_516"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_514_cast_fp16 = transpose(perm = var_514_perm_0, x = attn_output_15_cast_fp16)[name = tensor<string, []>("transpose_288")]; |
| tensor<fp16, [1, 1500, 1280]> input_77_cast_fp16 = reshape(shape = var_516, x = var_514_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299975616)))]; |
| tensor<fp16, [1280]> encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303252480)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_45_cast_fp16 = linear(bias = encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_out_proj_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_79_cast_fp16 = add(x = input_75_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| tensor<int32, [1]> input_81_axes_0 = const()[name = tensor<string, []>("input_81_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_7_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303255104)))]; |
| tensor<fp16, [1280]> encoder_layers_7_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303257728)))]; |
| tensor<fp16, [1, 1500, 1280]> input_81_cast_fp16 = layer_norm(axes = input_81_axes_0, beta = encoder_layers_7_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_7_final_layer_norm_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_7_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303260352)))]; |
| tensor<fp16, [5120]> encoder_layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316367616)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_46_cast_fp16 = linear(bias = encoder_layers_7_fc1_bias_to_fp16, weight = encoder_layers_7_fc1_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; |
| tensor<string, []> input_83_mode_0 = const()[name = tensor<string, []>("input_83_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_83_cast_fp16 = gelu(mode = input_83_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_7_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316377920)))]; |
| tensor<fp16, [1280]> encoder_layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329485184)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_47_cast_fp16 = linear(bias = encoder_layers_7_fc2_bias_to_fp16, weight = encoder_layers_7_fc2_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_85_cast_fp16 = add(x = input_79_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_53_axes_0 = const()[name = tensor<string, []>("hidden_states_53_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_8_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329487808)))]; |
| tensor<fp16, [1280]> encoder_layers_8_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329490432)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_53_cast_fp16 = layer_norm(axes = hidden_states_53_axes_0, beta = encoder_layers_8_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_8_self_attn_layer_norm_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("hidden_states_53_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329493056)))]; |
| tensor<fp16, [1280]> encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332769920)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_48_cast_fp16 = linear(bias = encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332772544)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336049408)))]; |
| tensor<fp16, [1280]> encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339326272)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_50_cast_fp16 = linear(bias = encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_states_53_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")]; |
| tensor<int32, [4]> var_557 = const()[name = tensor<string, []>("op_557"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_558_cast_fp16 = reshape(shape = var_557, x = linear_48_cast_fp16)[name = tensor<string, []>("op_558_cast_fp16")]; |
| tensor<int32, [4]> var_560 = const()[name = tensor<string, []>("op_560"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_561_cast_fp16 = reshape(shape = var_560, x = linear_49_cast_fp16)[name = tensor<string, []>("op_561_cast_fp16")]; |
| tensor<int32, [4]> var_563 = const()[name = tensor<string, []>("op_563"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_564_cast_fp16 = reshape(shape = var_563, x = linear_50_cast_fp16)[name = tensor<string, []>("op_564_cast_fp16")]; |
| tensor<int32, [4]> value_35_perm_0 = const()[name = tensor<string, []>("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_8_cast_fp16 = mul(x = var_558_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_8_cast_fp16")]; |
| tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_144_perm_0 = const()[name = tensor<string, []>("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_145_perm_0 = const()[name = tensor<string, []>("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_145 = transpose(perm = transpose_145_perm_0, x = var_561_cast_fp16)[name = tensor<string, []>("transpose_285")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_144 = transpose(perm = transpose_144_perm_0, x = mul_8_cast_fp16)[name = tensor<string, []>("transpose_286")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_144, y = transpose_145)[name = tensor<string, []>("matmul_8_cast_fp16")]; |
| tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = matmul_8_cast_fp16)[name = tensor<string, []>("softmax_8_cast_fp16")]; |
| tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = var_564_cast_fp16)[name = tensor<string, []>("transpose_287")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_8_cast_fp16, y = value_35_cast_fp16)[name = tensor<string, []>("attn_output_17_cast_fp16")]; |
| tensor<int32, [4]> var_567_perm_0 = const()[name = tensor<string, []>("op_567_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_567_cast_fp16 = transpose(perm = var_567_perm_0, x = attn_output_17_cast_fp16)[name = tensor<string, []>("transpose_284")]; |
| tensor<fp16, [1, 1500, 1280]> input_87_cast_fp16 = reshape(shape = var_569, x = var_567_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339328896)))]; |
| tensor<fp16, [1280]> encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342605760)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_51_cast_fp16 = linear(bias = encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_out_proj_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_89_cast_fp16 = add(x = input_85_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; |
| tensor<int32, [1]> input_91_axes_0 = const()[name = tensor<string, []>("input_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_8_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342608384)))]; |
| tensor<fp16, [1280]> encoder_layers_8_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342611008)))]; |
| tensor<fp16, [1, 1500, 1280]> input_91_cast_fp16 = layer_norm(axes = input_91_axes_0, beta = encoder_layers_8_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_8_final_layer_norm_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_8_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342613632)))]; |
| tensor<fp16, [5120]> encoder_layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355720896)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_52_cast_fp16 = linear(bias = encoder_layers_8_fc1_bias_to_fp16, weight = encoder_layers_8_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; |
| tensor<string, []> input_93_mode_0 = const()[name = tensor<string, []>("input_93_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_93_cast_fp16 = gelu(mode = input_93_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_8_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355731200)))]; |
| tensor<fp16, [1280]> encoder_layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368838464)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_53_cast_fp16 = linear(bias = encoder_layers_8_fc2_bias_to_fp16, weight = encoder_layers_8_fc2_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_95_cast_fp16 = add(x = input_89_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_59_axes_0 = const()[name = tensor<string, []>("hidden_states_59_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_9_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368841088)))]; |
| tensor<fp16, [1280]> encoder_layers_9_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368843712)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_59_cast_fp16 = layer_norm(axes = hidden_states_59_axes_0, beta = encoder_layers_9_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_9_self_attn_layer_norm_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("hidden_states_59_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368846336)))]; |
| tensor<fp16, [1280]> encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372123200)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_54_cast_fp16 = linear(bias = encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372125824)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375402688)))]; |
| tensor<fp16, [1280]> encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378679552)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_56_cast_fp16 = linear(bias = encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")]; |
| tensor<int32, [4]> var_610 = const()[name = tensor<string, []>("op_610"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_611_cast_fp16 = reshape(shape = var_610, x = linear_54_cast_fp16)[name = tensor<string, []>("op_611_cast_fp16")]; |
| tensor<int32, [4]> var_613 = const()[name = tensor<string, []>("op_613"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_614_cast_fp16 = reshape(shape = var_613, x = linear_55_cast_fp16)[name = tensor<string, []>("op_614_cast_fp16")]; |
| tensor<int32, [4]> var_616 = const()[name = tensor<string, []>("op_616"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_617_cast_fp16 = reshape(shape = var_616, x = linear_56_cast_fp16)[name = tensor<string, []>("op_617_cast_fp16")]; |
| tensor<int32, [4]> value_39_perm_0 = const()[name = tensor<string, []>("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_9_cast_fp16 = mul(x = var_611_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_9_cast_fp16")]; |
| tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_146_perm_0 = const()[name = tensor<string, []>("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_147_perm_0 = const()[name = tensor<string, []>("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_147 = transpose(perm = transpose_147_perm_0, x = var_614_cast_fp16)[name = tensor<string, []>("transpose_281")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_146 = transpose(perm = transpose_146_perm_0, x = mul_9_cast_fp16)[name = tensor<string, []>("transpose_282")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_146, y = transpose_147)[name = tensor<string, []>("matmul_9_cast_fp16")]; |
| tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = matmul_9_cast_fp16)[name = tensor<string, []>("softmax_9_cast_fp16")]; |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = var_617_cast_fp16)[name = tensor<string, []>("transpose_283")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = softmax_9_cast_fp16, y = value_39_cast_fp16)[name = tensor<string, []>("attn_output_19_cast_fp16")]; |
| tensor<int32, [4]> var_620_perm_0 = const()[name = tensor<string, []>("op_620_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_622 = const()[name = tensor<string, []>("op_622"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_620_cast_fp16 = transpose(perm = var_620_perm_0, x = attn_output_19_cast_fp16)[name = tensor<string, []>("transpose_280")]; |
| tensor<fp16, [1, 1500, 1280]> input_97_cast_fp16 = reshape(shape = var_622, x = var_620_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378682176)))]; |
| tensor<fp16, [1280]> encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381959040)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_57_cast_fp16 = linear(bias = encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_out_proj_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_99_cast_fp16 = add(x = input_95_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; |
| tensor<int32, [1]> input_101_axes_0 = const()[name = tensor<string, []>("input_101_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_9_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381961664)))]; |
| tensor<fp16, [1280]> encoder_layers_9_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381964288)))]; |
| tensor<fp16, [1, 1500, 1280]> input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = encoder_layers_9_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_9_final_layer_norm_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_9_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381966912)))]; |
| tensor<fp16, [5120]> encoder_layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395074176)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_58_cast_fp16 = linear(bias = encoder_layers_9_fc1_bias_to_fp16, weight = encoder_layers_9_fc1_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; |
| tensor<string, []> input_103_mode_0 = const()[name = tensor<string, []>("input_103_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_9_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395084480)))]; |
| tensor<fp16, [1280]> encoder_layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408191744)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_59_cast_fp16 = linear(bias = encoder_layers_9_fc2_bias_to_fp16, weight = encoder_layers_9_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_105_cast_fp16 = add(x = input_99_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_65_axes_0 = const()[name = tensor<string, []>("hidden_states_65_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_10_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408194368)))]; |
| tensor<fp16, [1280]> encoder_layers_10_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408196992)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = encoder_layers_10_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_10_self_attn_layer_norm_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("hidden_states_65_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408199616)))]; |
| tensor<fp16, [1280]> encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411476480)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_60_cast_fp16 = linear(bias = encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411479104)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414755968)))]; |
| tensor<fp16, [1280]> encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418032832)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_62_cast_fp16 = linear(bias = encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")]; |
| tensor<int32, [4]> var_663 = const()[name = tensor<string, []>("op_663"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_664_cast_fp16 = reshape(shape = var_663, x = linear_60_cast_fp16)[name = tensor<string, []>("op_664_cast_fp16")]; |
| tensor<int32, [4]> var_666 = const()[name = tensor<string, []>("op_666"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_667_cast_fp16 = reshape(shape = var_666, x = linear_61_cast_fp16)[name = tensor<string, []>("op_667_cast_fp16")]; |
| tensor<int32, [4]> var_669 = const()[name = tensor<string, []>("op_669"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_670_cast_fp16 = reshape(shape = var_669, x = linear_62_cast_fp16)[name = tensor<string, []>("op_670_cast_fp16")]; |
| tensor<int32, [4]> value_43_perm_0 = const()[name = tensor<string, []>("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_10_cast_fp16 = mul(x = var_664_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_10_cast_fp16")]; |
| tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_148_perm_0 = const()[name = tensor<string, []>("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_149_perm_0 = const()[name = tensor<string, []>("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_149 = transpose(perm = transpose_149_perm_0, x = var_667_cast_fp16)[name = tensor<string, []>("transpose_277")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_148 = transpose(perm = transpose_148_perm_0, x = mul_10_cast_fp16)[name = tensor<string, []>("transpose_278")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_148, y = transpose_149)[name = tensor<string, []>("matmul_10_cast_fp16")]; |
| tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = matmul_10_cast_fp16)[name = tensor<string, []>("softmax_10_cast_fp16")]; |
| tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = var_670_cast_fp16)[name = tensor<string, []>("transpose_279")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_10_cast_fp16, y = value_43_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; |
| tensor<int32, [4]> var_673_perm_0 = const()[name = tensor<string, []>("op_673_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_675 = const()[name = tensor<string, []>("op_675"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_673_cast_fp16 = transpose(perm = var_673_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_276")]; |
| tensor<fp16, [1, 1500, 1280]> input_107_cast_fp16 = reshape(shape = var_675, x = var_673_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418035456)))]; |
| tensor<fp16, [1280]> encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421312320)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_63_cast_fp16 = linear(bias = encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_out_proj_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_109_cast_fp16 = add(x = input_105_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; |
| tensor<int32, [1]> input_111_axes_0 = const()[name = tensor<string, []>("input_111_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_10_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421314944)))]; |
| tensor<fp16, [1280]> encoder_layers_10_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421317568)))]; |
| tensor<fp16, [1, 1500, 1280]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = encoder_layers_10_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_10_final_layer_norm_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_10_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421320192)))]; |
| tensor<fp16, [5120]> encoder_layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434427456)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_64_cast_fp16 = linear(bias = encoder_layers_10_fc1_bias_to_fp16, weight = encoder_layers_10_fc1_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; |
| tensor<string, []> input_113_mode_0 = const()[name = tensor<string, []>("input_113_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_113_cast_fp16 = gelu(mode = input_113_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_10_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434437760)))]; |
| tensor<fp16, [1280]> encoder_layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447545024)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_65_cast_fp16 = linear(bias = encoder_layers_10_fc2_bias_to_fp16, weight = encoder_layers_10_fc2_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_115_cast_fp16 = add(x = input_109_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_71_axes_0 = const()[name = tensor<string, []>("hidden_states_71_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_11_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447547648)))]; |
| tensor<fp16, [1280]> encoder_layers_11_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447550272)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_71_cast_fp16 = layer_norm(axes = hidden_states_71_axes_0, beta = encoder_layers_11_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_11_self_attn_layer_norm_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("hidden_states_71_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447552896)))]; |
| tensor<fp16, [1280]> encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450829760)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_66_cast_fp16 = linear(bias = encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450832384)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454109248)))]; |
| tensor<fp16, [1280]> encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457386112)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_68_cast_fp16 = linear(bias = encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")]; |
| tensor<int32, [4]> var_716 = const()[name = tensor<string, []>("op_716"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_717_cast_fp16 = reshape(shape = var_716, x = linear_66_cast_fp16)[name = tensor<string, []>("op_717_cast_fp16")]; |
| tensor<int32, [4]> var_719 = const()[name = tensor<string, []>("op_719"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_720_cast_fp16 = reshape(shape = var_719, x = linear_67_cast_fp16)[name = tensor<string, []>("op_720_cast_fp16")]; |
| tensor<int32, [4]> var_722 = const()[name = tensor<string, []>("op_722"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_723_cast_fp16 = reshape(shape = var_722, x = linear_68_cast_fp16)[name = tensor<string, []>("op_723_cast_fp16")]; |
| tensor<int32, [4]> value_47_perm_0 = const()[name = tensor<string, []>("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_11_cast_fp16 = mul(x = var_717_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_11_cast_fp16")]; |
| tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_150_perm_0 = const()[name = tensor<string, []>("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_151_perm_0 = const()[name = tensor<string, []>("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_151 = transpose(perm = transpose_151_perm_0, x = var_720_cast_fp16)[name = tensor<string, []>("transpose_273")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_150 = transpose(perm = transpose_150_perm_0, x = mul_11_cast_fp16)[name = tensor<string, []>("transpose_274")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_150, y = transpose_151)[name = tensor<string, []>("matmul_11_cast_fp16")]; |
| tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = matmul_11_cast_fp16)[name = tensor<string, []>("softmax_11_cast_fp16")]; |
| tensor<bool, []> attn_output_23_transpose_x_0 = const()[name = tensor<string, []>("attn_output_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_23_transpose_y_0 = const()[name = tensor<string, []>("attn_output_23_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = var_723_cast_fp16)[name = tensor<string, []>("transpose_275")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_23_cast_fp16 = matmul(transpose_x = attn_output_23_transpose_x_0, transpose_y = attn_output_23_transpose_y_0, x = softmax_11_cast_fp16, y = value_47_cast_fp16)[name = tensor<string, []>("attn_output_23_cast_fp16")]; |
| tensor<int32, [4]> var_726_perm_0 = const()[name = tensor<string, []>("op_726_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_728 = const()[name = tensor<string, []>("op_728"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_726_cast_fp16 = transpose(perm = var_726_perm_0, x = attn_output_23_cast_fp16)[name = tensor<string, []>("transpose_272")]; |
| tensor<fp16, [1, 1500, 1280]> input_117_cast_fp16 = reshape(shape = var_728, x = var_726_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457388736)))]; |
| tensor<fp16, [1280]> encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460665600)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_69_cast_fp16 = linear(bias = encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_out_proj_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_119_cast_fp16 = add(x = input_115_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; |
| tensor<int32, [1]> input_121_axes_0 = const()[name = tensor<string, []>("input_121_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_11_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460668224)))]; |
| tensor<fp16, [1280]> encoder_layers_11_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460670848)))]; |
| tensor<fp16, [1, 1500, 1280]> input_121_cast_fp16 = layer_norm(axes = input_121_axes_0, beta = encoder_layers_11_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_11_final_layer_norm_weight_to_fp16, x = input_119_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_11_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460673472)))]; |
| tensor<fp16, [5120]> encoder_layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473780736)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_70_cast_fp16 = linear(bias = encoder_layers_11_fc1_bias_to_fp16, weight = encoder_layers_11_fc1_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; |
| tensor<string, []> input_123_mode_0 = const()[name = tensor<string, []>("input_123_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_123_cast_fp16 = gelu(mode = input_123_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_11_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473791040)))]; |
| tensor<fp16, [1280]> encoder_layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486898304)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_71_cast_fp16 = linear(bias = encoder_layers_11_fc2_bias_to_fp16, weight = encoder_layers_11_fc2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_125_cast_fp16 = add(x = input_119_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_77_axes_0 = const()[name = tensor<string, []>("hidden_states_77_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_12_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486900928)))]; |
| tensor<fp16, [1280]> encoder_layers_12_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486903552)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_77_cast_fp16 = layer_norm(axes = hidden_states_77_axes_0, beta = encoder_layers_12_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_12_self_attn_layer_norm_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("hidden_states_77_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486906176)))]; |
| tensor<fp16, [1280]> encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490183040)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_72_cast_fp16 = linear(bias = encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490185664)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493462528)))]; |
| tensor<fp16, [1280]> encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496739392)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_74_cast_fp16 = linear(bias = encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")]; |
| tensor<int32, [4]> var_769 = const()[name = tensor<string, []>("op_769"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_770_cast_fp16 = reshape(shape = var_769, x = linear_72_cast_fp16)[name = tensor<string, []>("op_770_cast_fp16")]; |
| tensor<int32, [4]> var_772 = const()[name = tensor<string, []>("op_772"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_773_cast_fp16 = reshape(shape = var_772, x = linear_73_cast_fp16)[name = tensor<string, []>("op_773_cast_fp16")]; |
| tensor<int32, [4]> var_775 = const()[name = tensor<string, []>("op_775"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_776_cast_fp16 = reshape(shape = var_775, x = linear_74_cast_fp16)[name = tensor<string, []>("op_776_cast_fp16")]; |
| tensor<int32, [4]> value_51_perm_0 = const()[name = tensor<string, []>("value_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_12_cast_fp16 = mul(x = var_770_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_12_cast_fp16")]; |
| tensor<bool, []> matmul_12_transpose_y_0 = const()[name = tensor<string, []>("matmul_12_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_12_transpose_x_0 = const()[name = tensor<string, []>("matmul_12_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_152_perm_0 = const()[name = tensor<string, []>("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_153_perm_0 = const()[name = tensor<string, []>("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_153 = transpose(perm = transpose_153_perm_0, x = var_773_cast_fp16)[name = tensor<string, []>("transpose_269")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_152 = transpose(perm = transpose_152_perm_0, x = mul_12_cast_fp16)[name = tensor<string, []>("transpose_270")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_152, y = transpose_153)[name = tensor<string, []>("matmul_12_cast_fp16")]; |
| tensor<int32, []> softmax_12_axis_0 = const()[name = tensor<string, []>("softmax_12_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_12_cast_fp16 = softmax(axis = softmax_12_axis_0, x = matmul_12_cast_fp16)[name = tensor<string, []>("softmax_12_cast_fp16")]; |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_51_cast_fp16 = transpose(perm = value_51_perm_0, x = var_776_cast_fp16)[name = tensor<string, []>("transpose_271")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_12_cast_fp16, y = value_51_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; |
| tensor<int32, [4]> var_779_perm_0 = const()[name = tensor<string, []>("op_779_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_781 = const()[name = tensor<string, []>("op_781"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_779_cast_fp16 = transpose(perm = var_779_perm_0, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_268")]; |
| tensor<fp16, [1, 1500, 1280]> input_127_cast_fp16 = reshape(shape = var_781, x = var_779_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496742016)))]; |
| tensor<fp16, [1280]> encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500018880)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_75_cast_fp16 = linear(bias = encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_out_proj_weight_to_fp16, x = input_127_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_129_cast_fp16 = add(x = input_125_cast_fp16, y = linear_75_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; |
| tensor<int32, [1]> input_131_axes_0 = const()[name = tensor<string, []>("input_131_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_12_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500021504)))]; |
| tensor<fp16, [1280]> encoder_layers_12_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500024128)))]; |
| tensor<fp16, [1, 1500, 1280]> input_131_cast_fp16 = layer_norm(axes = input_131_axes_0, beta = encoder_layers_12_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_12_final_layer_norm_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_12_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500026752)))]; |
| tensor<fp16, [5120]> encoder_layers_12_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513134016)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_76_cast_fp16 = linear(bias = encoder_layers_12_fc1_bias_to_fp16, weight = encoder_layers_12_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")]; |
| tensor<string, []> input_133_mode_0 = const()[name = tensor<string, []>("input_133_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = linear_76_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_12_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513144320)))]; |
| tensor<fp16, [1280]> encoder_layers_12_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526251584)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_77_cast_fp16 = linear(bias = encoder_layers_12_fc2_bias_to_fp16, weight = encoder_layers_12_fc2_weight_to_fp16, x = input_133_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_135_cast_fp16 = add(x = input_129_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_83_axes_0 = const()[name = tensor<string, []>("hidden_states_83_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_13_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526254208)))]; |
| tensor<fp16, [1280]> encoder_layers_13_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526256832)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_83_cast_fp16 = layer_norm(axes = hidden_states_83_axes_0, beta = encoder_layers_13_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_13_self_attn_layer_norm_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("hidden_states_83_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526259456)))]; |
| tensor<fp16, [1280]> encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529536320)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_78_cast_fp16 = linear(bias = encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529538944)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532815808)))]; |
| tensor<fp16, [1280]> encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536092672)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_80_cast_fp16 = linear(bias = encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")]; |
| tensor<int32, [4]> var_822 = const()[name = tensor<string, []>("op_822"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_823_cast_fp16 = reshape(shape = var_822, x = linear_78_cast_fp16)[name = tensor<string, []>("op_823_cast_fp16")]; |
| tensor<int32, [4]> var_825 = const()[name = tensor<string, []>("op_825"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_826_cast_fp16 = reshape(shape = var_825, x = linear_79_cast_fp16)[name = tensor<string, []>("op_826_cast_fp16")]; |
| tensor<int32, [4]> var_828 = const()[name = tensor<string, []>("op_828"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_829_cast_fp16 = reshape(shape = var_828, x = linear_80_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")]; |
| tensor<int32, [4]> value_55_perm_0 = const()[name = tensor<string, []>("value_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_13_cast_fp16 = mul(x = var_823_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_13_cast_fp16")]; |
| tensor<bool, []> matmul_13_transpose_y_0 = const()[name = tensor<string, []>("matmul_13_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_13_transpose_x_0 = const()[name = tensor<string, []>("matmul_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_154_perm_0 = const()[name = tensor<string, []>("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_155_perm_0 = const()[name = tensor<string, []>("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_155 = transpose(perm = transpose_155_perm_0, x = var_826_cast_fp16)[name = tensor<string, []>("transpose_265")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_154 = transpose(perm = transpose_154_perm_0, x = mul_13_cast_fp16)[name = tensor<string, []>("transpose_266")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = transpose_154, y = transpose_155)[name = tensor<string, []>("matmul_13_cast_fp16")]; |
| tensor<int32, []> softmax_13_axis_0 = const()[name = tensor<string, []>("softmax_13_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_13_cast_fp16 = softmax(axis = softmax_13_axis_0, x = matmul_13_cast_fp16)[name = tensor<string, []>("softmax_13_cast_fp16")]; |
| tensor<bool, []> attn_output_27_transpose_x_0 = const()[name = tensor<string, []>("attn_output_27_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_27_transpose_y_0 = const()[name = tensor<string, []>("attn_output_27_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_55_cast_fp16 = transpose(perm = value_55_perm_0, x = var_829_cast_fp16)[name = tensor<string, []>("transpose_267")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_27_cast_fp16 = matmul(transpose_x = attn_output_27_transpose_x_0, transpose_y = attn_output_27_transpose_y_0, x = softmax_13_cast_fp16, y = value_55_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; |
| tensor<int32, [4]> var_832_perm_0 = const()[name = tensor<string, []>("op_832_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_834 = const()[name = tensor<string, []>("op_834"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_832_cast_fp16 = transpose(perm = var_832_perm_0, x = attn_output_27_cast_fp16)[name = tensor<string, []>("transpose_264")]; |
| tensor<fp16, [1, 1500, 1280]> input_137_cast_fp16 = reshape(shape = var_834, x = var_832_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536095296)))]; |
| tensor<fp16, [1280]> encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539372160)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_81_cast_fp16 = linear(bias = encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_out_proj_weight_to_fp16, x = input_137_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_139_cast_fp16 = add(x = input_135_cast_fp16, y = linear_81_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_13_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539374784)))]; |
| tensor<fp16, [1280]> encoder_layers_13_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539377408)))]; |
| tensor<fp16, [1, 1500, 1280]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = encoder_layers_13_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_13_final_layer_norm_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_13_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539380032)))]; |
| tensor<fp16, [5120]> encoder_layers_13_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552487296)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_82_cast_fp16 = linear(bias = encoder_layers_13_fc1_bias_to_fp16, weight = encoder_layers_13_fc1_weight_to_fp16, x = input_141_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")]; |
| tensor<string, []> input_143_mode_0 = const()[name = tensor<string, []>("input_143_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = linear_82_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_13_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552497600)))]; |
| tensor<fp16, [1280]> encoder_layers_13_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565604864)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_83_cast_fp16 = linear(bias = encoder_layers_13_fc2_bias_to_fp16, weight = encoder_layers_13_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_145_cast_fp16 = add(x = input_139_cast_fp16, y = linear_83_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_89_axes_0 = const()[name = tensor<string, []>("hidden_states_89_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_14_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565607488)))]; |
| tensor<fp16, [1280]> encoder_layers_14_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565610112)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_89_cast_fp16 = layer_norm(axes = hidden_states_89_axes_0, beta = encoder_layers_14_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_14_self_attn_layer_norm_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("hidden_states_89_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565612736)))]; |
| tensor<fp16, [1280]> encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568889600)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_84_cast_fp16 = linear(bias = encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568892224)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572169088)))]; |
| tensor<fp16, [1280]> encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575445952)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_86_cast_fp16 = linear(bias = encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")]; |
| tensor<int32, [4]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_876_cast_fp16 = reshape(shape = var_875, x = linear_84_cast_fp16)[name = tensor<string, []>("op_876_cast_fp16")]; |
| tensor<int32, [4]> var_878 = const()[name = tensor<string, []>("op_878"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_879_cast_fp16 = reshape(shape = var_878, x = linear_85_cast_fp16)[name = tensor<string, []>("op_879_cast_fp16")]; |
| tensor<int32, [4]> var_881 = const()[name = tensor<string, []>("op_881"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_882_cast_fp16 = reshape(shape = var_881, x = linear_86_cast_fp16)[name = tensor<string, []>("op_882_cast_fp16")]; |
| tensor<int32, [4]> value_59_perm_0 = const()[name = tensor<string, []>("value_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_14_cast_fp16 = mul(x = var_876_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_14_cast_fp16")]; |
| tensor<bool, []> matmul_14_transpose_y_0 = const()[name = tensor<string, []>("matmul_14_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_14_transpose_x_0 = const()[name = tensor<string, []>("matmul_14_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_156_perm_0 = const()[name = tensor<string, []>("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_157_perm_0 = const()[name = tensor<string, []>("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_157 = transpose(perm = transpose_157_perm_0, x = var_879_cast_fp16)[name = tensor<string, []>("transpose_261")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_156 = transpose(perm = transpose_156_perm_0, x = mul_14_cast_fp16)[name = tensor<string, []>("transpose_262")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_156, y = transpose_157)[name = tensor<string, []>("matmul_14_cast_fp16")]; |
| tensor<int32, []> softmax_14_axis_0 = const()[name = tensor<string, []>("softmax_14_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_14_cast_fp16 = softmax(axis = softmax_14_axis_0, x = matmul_14_cast_fp16)[name = tensor<string, []>("softmax_14_cast_fp16")]; |
| tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_59_cast_fp16 = transpose(perm = value_59_perm_0, x = var_882_cast_fp16)[name = tensor<string, []>("transpose_263")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_14_cast_fp16, y = value_59_cast_fp16)[name = tensor<string, []>("attn_output_29_cast_fp16")]; |
| tensor<int32, [4]> var_885_perm_0 = const()[name = tensor<string, []>("op_885_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_887 = const()[name = tensor<string, []>("op_887"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_885_cast_fp16 = transpose(perm = var_885_perm_0, x = attn_output_29_cast_fp16)[name = tensor<string, []>("transpose_260")]; |
| tensor<fp16, [1, 1500, 1280]> input_147_cast_fp16 = reshape(shape = var_887, x = var_885_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575448576)))]; |
| tensor<fp16, [1280]> encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578725440)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_87_cast_fp16 = linear(bias = encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_out_proj_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_149_cast_fp16 = add(x = input_145_cast_fp16, y = linear_87_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; |
| tensor<int32, [1]> input_151_axes_0 = const()[name = tensor<string, []>("input_151_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_14_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578728064)))]; |
| tensor<fp16, [1280]> encoder_layers_14_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578730688)))]; |
| tensor<fp16, [1, 1500, 1280]> input_151_cast_fp16 = layer_norm(axes = input_151_axes_0, beta = encoder_layers_14_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_14_final_layer_norm_weight_to_fp16, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_14_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578733312)))]; |
| tensor<fp16, [5120]> encoder_layers_14_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591840576)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_88_cast_fp16 = linear(bias = encoder_layers_14_fc1_bias_to_fp16, weight = encoder_layers_14_fc1_weight_to_fp16, x = input_151_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")]; |
| tensor<string, []> input_153_mode_0 = const()[name = tensor<string, []>("input_153_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_153_cast_fp16 = gelu(mode = input_153_mode_0, x = linear_88_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_14_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591850880)))]; |
| tensor<fp16, [1280]> encoder_layers_14_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604958144)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_89_cast_fp16 = linear(bias = encoder_layers_14_fc2_bias_to_fp16, weight = encoder_layers_14_fc2_weight_to_fp16, x = input_153_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_155_cast_fp16 = add(x = input_149_cast_fp16, y = linear_89_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_95_axes_0 = const()[name = tensor<string, []>("hidden_states_95_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_15_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604960768)))]; |
| tensor<fp16, [1280]> encoder_layers_15_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604963392)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_95_cast_fp16 = layer_norm(axes = hidden_states_95_axes_0, beta = encoder_layers_15_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_15_self_attn_layer_norm_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("hidden_states_95_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604966016)))]; |
| tensor<fp16, [1280]> encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608242880)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_90_cast_fp16 = linear(bias = encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608245504)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611522368)))]; |
| tensor<fp16, [1280]> encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614799232)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_92_cast_fp16 = linear(bias = encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")]; |
| tensor<int32, [4]> var_928 = const()[name = tensor<string, []>("op_928"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_929_cast_fp16 = reshape(shape = var_928, x = linear_90_cast_fp16)[name = tensor<string, []>("op_929_cast_fp16")]; |
| tensor<int32, [4]> var_931 = const()[name = tensor<string, []>("op_931"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_932_cast_fp16 = reshape(shape = var_931, x = linear_91_cast_fp16)[name = tensor<string, []>("op_932_cast_fp16")]; |
| tensor<int32, [4]> var_934 = const()[name = tensor<string, []>("op_934"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_935_cast_fp16 = reshape(shape = var_934, x = linear_92_cast_fp16)[name = tensor<string, []>("op_935_cast_fp16")]; |
| tensor<int32, [4]> value_63_perm_0 = const()[name = tensor<string, []>("value_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_15_cast_fp16 = mul(x = var_929_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_15_cast_fp16")]; |
| tensor<bool, []> matmul_15_transpose_y_0 = const()[name = tensor<string, []>("matmul_15_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_15_transpose_x_0 = const()[name = tensor<string, []>("matmul_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_158_perm_0 = const()[name = tensor<string, []>("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_159_perm_0 = const()[name = tensor<string, []>("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_159 = transpose(perm = transpose_159_perm_0, x = var_932_cast_fp16)[name = tensor<string, []>("transpose_257")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_158 = transpose(perm = transpose_158_perm_0, x = mul_15_cast_fp16)[name = tensor<string, []>("transpose_258")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = transpose_158, y = transpose_159)[name = tensor<string, []>("matmul_15_cast_fp16")]; |
| tensor<int32, []> softmax_15_axis_0 = const()[name = tensor<string, []>("softmax_15_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_15_cast_fp16 = softmax(axis = softmax_15_axis_0, x = matmul_15_cast_fp16)[name = tensor<string, []>("softmax_15_cast_fp16")]; |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_63_cast_fp16 = transpose(perm = value_63_perm_0, x = var_935_cast_fp16)[name = tensor<string, []>("transpose_259")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = softmax_15_cast_fp16, y = value_63_cast_fp16)[name = tensor<string, []>("attn_output_31_cast_fp16")]; |
| tensor<int32, [4]> var_938_perm_0 = const()[name = tensor<string, []>("op_938_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_940 = const()[name = tensor<string, []>("op_940"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_938_cast_fp16 = transpose(perm = var_938_perm_0, x = attn_output_31_cast_fp16)[name = tensor<string, []>("transpose_256")]; |
| tensor<fp16, [1, 1500, 1280]> input_157_cast_fp16 = reshape(shape = var_940, x = var_938_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614801856)))]; |
| tensor<fp16, [1280]> encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618078720)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_93_cast_fp16 = linear(bias = encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_out_proj_weight_to_fp16, x = input_157_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_159_cast_fp16 = add(x = input_155_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")]; |
| tensor<int32, [1]> input_161_axes_0 = const()[name = tensor<string, []>("input_161_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_15_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618081344)))]; |
| tensor<fp16, [1280]> encoder_layers_15_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618083968)))]; |
| tensor<fp16, [1, 1500, 1280]> input_161_cast_fp16 = layer_norm(axes = input_161_axes_0, beta = encoder_layers_15_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_15_final_layer_norm_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_15_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618086592)))]; |
| tensor<fp16, [5120]> encoder_layers_15_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631193856)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_94_cast_fp16 = linear(bias = encoder_layers_15_fc1_bias_to_fp16, weight = encoder_layers_15_fc1_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")]; |
| tensor<string, []> input_163_mode_0 = const()[name = tensor<string, []>("input_163_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = linear_94_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_15_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631204160)))]; |
| tensor<fp16, [1280]> encoder_layers_15_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644311424)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_95_cast_fp16 = linear(bias = encoder_layers_15_fc2_bias_to_fp16, weight = encoder_layers_15_fc2_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_165_cast_fp16 = add(x = input_159_cast_fp16, y = linear_95_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_101_axes_0 = const()[name = tensor<string, []>("hidden_states_101_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_16_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644314048)))]; |
| tensor<fp16, [1280]> encoder_layers_16_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644316672)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_101_cast_fp16 = layer_norm(axes = hidden_states_101_axes_0, beta = encoder_layers_16_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_16_self_attn_layer_norm_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("hidden_states_101_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644319296)))]; |
| tensor<fp16, [1280]> encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647596160)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_96_cast_fp16 = linear(bias = encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647598784)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650875648)))]; |
| tensor<fp16, [1280]> encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654152512)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_98_cast_fp16 = linear(bias = encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")]; |
| tensor<int32, [4]> var_981 = const()[name = tensor<string, []>("op_981"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_982_cast_fp16 = reshape(shape = var_981, x = linear_96_cast_fp16)[name = tensor<string, []>("op_982_cast_fp16")]; |
| tensor<int32, [4]> var_984 = const()[name = tensor<string, []>("op_984"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_985_cast_fp16 = reshape(shape = var_984, x = linear_97_cast_fp16)[name = tensor<string, []>("op_985_cast_fp16")]; |
| tensor<int32, [4]> var_987 = const()[name = tensor<string, []>("op_987"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_988_cast_fp16 = reshape(shape = var_987, x = linear_98_cast_fp16)[name = tensor<string, []>("op_988_cast_fp16")]; |
| tensor<int32, [4]> value_67_perm_0 = const()[name = tensor<string, []>("value_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_16_cast_fp16 = mul(x = var_982_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_16_cast_fp16")]; |
| tensor<bool, []> matmul_16_transpose_y_0 = const()[name = tensor<string, []>("matmul_16_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_16_transpose_x_0 = const()[name = tensor<string, []>("matmul_16_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_160_perm_0 = const()[name = tensor<string, []>("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_161_perm_0 = const()[name = tensor<string, []>("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_161 = transpose(perm = transpose_161_perm_0, x = var_985_cast_fp16)[name = tensor<string, []>("transpose_253")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_160 = transpose(perm = transpose_160_perm_0, x = mul_16_cast_fp16)[name = tensor<string, []>("transpose_254")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_160, y = transpose_161)[name = tensor<string, []>("matmul_16_cast_fp16")]; |
| tensor<int32, []> softmax_16_axis_0 = const()[name = tensor<string, []>("softmax_16_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_16_cast_fp16 = softmax(axis = softmax_16_axis_0, x = matmul_16_cast_fp16)[name = tensor<string, []>("softmax_16_cast_fp16")]; |
| tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_67_cast_fp16 = transpose(perm = value_67_perm_0, x = var_988_cast_fp16)[name = tensor<string, []>("transpose_255")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_16_cast_fp16, y = value_67_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; |
| tensor<int32, [4]> var_991_perm_0 = const()[name = tensor<string, []>("op_991_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_993 = const()[name = tensor<string, []>("op_993"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_991_cast_fp16 = transpose(perm = var_991_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_252")]; |
| tensor<fp16, [1, 1500, 1280]> input_167_cast_fp16 = reshape(shape = var_993, x = var_991_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654155136)))]; |
| tensor<fp16, [1280]> encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657432000)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_99_cast_fp16 = linear(bias = encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_out_proj_weight_to_fp16, x = input_167_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_169_cast_fp16 = add(x = input_165_cast_fp16, y = linear_99_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")]; |
| tensor<int32, [1]> input_171_axes_0 = const()[name = tensor<string, []>("input_171_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_16_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657434624)))]; |
| tensor<fp16, [1280]> encoder_layers_16_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657437248)))]; |
| tensor<fp16, [1, 1500, 1280]> input_171_cast_fp16 = layer_norm(axes = input_171_axes_0, beta = encoder_layers_16_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_16_final_layer_norm_weight_to_fp16, x = input_169_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_16_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657439872)))]; |
| tensor<fp16, [5120]> encoder_layers_16_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670547136)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_100_cast_fp16 = linear(bias = encoder_layers_16_fc1_bias_to_fp16, weight = encoder_layers_16_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")]; |
| tensor<string, []> input_173_mode_0 = const()[name = tensor<string, []>("input_173_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_173_cast_fp16 = gelu(mode = input_173_mode_0, x = linear_100_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_16_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670557440)))]; |
| tensor<fp16, [1280]> encoder_layers_16_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683664704)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_101_cast_fp16 = linear(bias = encoder_layers_16_fc2_bias_to_fp16, weight = encoder_layers_16_fc2_weight_to_fp16, x = input_173_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_175_cast_fp16 = add(x = input_169_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_107_axes_0 = const()[name = tensor<string, []>("hidden_states_107_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_17_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683667328)))]; |
| tensor<fp16, [1280]> encoder_layers_17_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683669952)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_107_cast_fp16 = layer_norm(axes = hidden_states_107_axes_0, beta = encoder_layers_17_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_17_self_attn_layer_norm_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("hidden_states_107_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683672576)))]; |
| tensor<fp16, [1280]> encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686949440)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_102_cast_fp16 = linear(bias = encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686952064)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690228928)))]; |
| tensor<fp16, [1280]> encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693505792)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_104_cast_fp16 = linear(bias = encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")]; |
| tensor<int32, [4]> var_1034 = const()[name = tensor<string, []>("op_1034"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1035_cast_fp16 = reshape(shape = var_1034, x = linear_102_cast_fp16)[name = tensor<string, []>("op_1035_cast_fp16")]; |
| tensor<int32, [4]> var_1037 = const()[name = tensor<string, []>("op_1037"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1038_cast_fp16 = reshape(shape = var_1037, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1038_cast_fp16")]; |
| tensor<int32, [4]> var_1040 = const()[name = tensor<string, []>("op_1040"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1041_cast_fp16 = reshape(shape = var_1040, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1041_cast_fp16")]; |
| tensor<int32, [4]> value_71_perm_0 = const()[name = tensor<string, []>("value_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_17_cast_fp16 = mul(x = var_1035_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_17_cast_fp16")]; |
| tensor<bool, []> matmul_17_transpose_y_0 = const()[name = tensor<string, []>("matmul_17_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_17_transpose_x_0 = const()[name = tensor<string, []>("matmul_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_162_perm_0 = const()[name = tensor<string, []>("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_163_perm_0 = const()[name = tensor<string, []>("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_163 = transpose(perm = transpose_163_perm_0, x = var_1038_cast_fp16)[name = tensor<string, []>("transpose_249")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_162 = transpose(perm = transpose_162_perm_0, x = mul_17_cast_fp16)[name = tensor<string, []>("transpose_250")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = transpose_162, y = transpose_163)[name = tensor<string, []>("matmul_17_cast_fp16")]; |
| tensor<int32, []> softmax_17_axis_0 = const()[name = tensor<string, []>("softmax_17_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_17_cast_fp16 = softmax(axis = softmax_17_axis_0, x = matmul_17_cast_fp16)[name = tensor<string, []>("softmax_17_cast_fp16")]; |
| tensor<bool, []> attn_output_35_transpose_x_0 = const()[name = tensor<string, []>("attn_output_35_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_35_transpose_y_0 = const()[name = tensor<string, []>("attn_output_35_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_71_cast_fp16 = transpose(perm = value_71_perm_0, x = var_1041_cast_fp16)[name = tensor<string, []>("transpose_251")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_35_cast_fp16 = matmul(transpose_x = attn_output_35_transpose_x_0, transpose_y = attn_output_35_transpose_y_0, x = softmax_17_cast_fp16, y = value_71_cast_fp16)[name = tensor<string, []>("attn_output_35_cast_fp16")]; |
| tensor<int32, [4]> var_1044_perm_0 = const()[name = tensor<string, []>("op_1044_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1046 = const()[name = tensor<string, []>("op_1046"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1044_cast_fp16 = transpose(perm = var_1044_perm_0, x = attn_output_35_cast_fp16)[name = tensor<string, []>("transpose_248")]; |
| tensor<fp16, [1, 1500, 1280]> input_177_cast_fp16 = reshape(shape = var_1046, x = var_1044_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693508416)))]; |
| tensor<fp16, [1280]> encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696785280)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_105_cast_fp16 = linear(bias = encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_out_proj_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_179_cast_fp16 = add(x = input_175_cast_fp16, y = linear_105_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; |
| tensor<int32, [1]> input_181_axes_0 = const()[name = tensor<string, []>("input_181_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_17_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696787904)))]; |
| tensor<fp16, [1280]> encoder_layers_17_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696790528)))]; |
| tensor<fp16, [1, 1500, 1280]> input_181_cast_fp16 = layer_norm(axes = input_181_axes_0, beta = encoder_layers_17_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_17_final_layer_norm_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_17_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696793152)))]; |
| tensor<fp16, [5120]> encoder_layers_17_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709900416)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_106_cast_fp16 = linear(bias = encoder_layers_17_fc1_bias_to_fp16, weight = encoder_layers_17_fc1_weight_to_fp16, x = input_181_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")]; |
| tensor<string, []> input_183_mode_0 = const()[name = tensor<string, []>("input_183_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = linear_106_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_17_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709910720)))]; |
| tensor<fp16, [1280]> encoder_layers_17_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723017984)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_107_cast_fp16 = linear(bias = encoder_layers_17_fc2_bias_to_fp16, weight = encoder_layers_17_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_185_cast_fp16 = add(x = input_179_cast_fp16, y = linear_107_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_113_axes_0 = const()[name = tensor<string, []>("hidden_states_113_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_18_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723020608)))]; |
| tensor<fp16, [1280]> encoder_layers_18_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723023232)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_113_cast_fp16 = layer_norm(axes = hidden_states_113_axes_0, beta = encoder_layers_18_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_18_self_attn_layer_norm_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("hidden_states_113_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723025856)))]; |
| tensor<fp16, [1280]> encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726302720)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_108_cast_fp16 = linear(bias = encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726305344)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729582208)))]; |
| tensor<fp16, [1280]> encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732859072)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_110_cast_fp16 = linear(bias = encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")]; |
| tensor<int32, [4]> var_1087 = const()[name = tensor<string, []>("op_1087"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1088_cast_fp16 = reshape(shape = var_1087, x = linear_108_cast_fp16)[name = tensor<string, []>("op_1088_cast_fp16")]; |
| tensor<int32, [4]> var_1090 = const()[name = tensor<string, []>("op_1090"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1091_cast_fp16 = reshape(shape = var_1090, x = linear_109_cast_fp16)[name = tensor<string, []>("op_1091_cast_fp16")]; |
| tensor<int32, [4]> var_1093 = const()[name = tensor<string, []>("op_1093"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1094_cast_fp16 = reshape(shape = var_1093, x = linear_110_cast_fp16)[name = tensor<string, []>("op_1094_cast_fp16")]; |
| tensor<int32, [4]> value_75_perm_0 = const()[name = tensor<string, []>("value_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_18_cast_fp16 = mul(x = var_1088_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_18_cast_fp16")]; |
| tensor<bool, []> matmul_18_transpose_y_0 = const()[name = tensor<string, []>("matmul_18_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_18_transpose_x_0 = const()[name = tensor<string, []>("matmul_18_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_164_perm_0 = const()[name = tensor<string, []>("transpose_164_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_165_perm_0 = const()[name = tensor<string, []>("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_165 = transpose(perm = transpose_165_perm_0, x = var_1091_cast_fp16)[name = tensor<string, []>("transpose_245")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_164 = transpose(perm = transpose_164_perm_0, x = mul_18_cast_fp16)[name = tensor<string, []>("transpose_246")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_164, y = transpose_165)[name = tensor<string, []>("matmul_18_cast_fp16")]; |
| tensor<int32, []> softmax_18_axis_0 = const()[name = tensor<string, []>("softmax_18_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_18_cast_fp16 = softmax(axis = softmax_18_axis_0, x = matmul_18_cast_fp16)[name = tensor<string, []>("softmax_18_cast_fp16")]; |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_75_cast_fp16 = transpose(perm = value_75_perm_0, x = var_1094_cast_fp16)[name = tensor<string, []>("transpose_247")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_18_cast_fp16, y = value_75_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; |
| tensor<int32, [4]> var_1097_perm_0 = const()[name = tensor<string, []>("op_1097_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1099 = const()[name = tensor<string, []>("op_1099"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1097_cast_fp16 = transpose(perm = var_1097_perm_0, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_244")]; |
| tensor<fp16, [1, 1500, 1280]> input_187_cast_fp16 = reshape(shape = var_1099, x = var_1097_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_18_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732861696)))]; |
| tensor<fp16, [1280]> encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736138560)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_111_cast_fp16 = linear(bias = encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_out_proj_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_189_cast_fp16 = add(x = input_185_cast_fp16, y = linear_111_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")]; |
| tensor<int32, [1]> input_191_axes_0 = const()[name = tensor<string, []>("input_191_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_18_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736141184)))]; |
| tensor<fp16, [1280]> encoder_layers_18_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736143808)))]; |
| tensor<fp16, [1, 1500, 1280]> input_191_cast_fp16 = layer_norm(axes = input_191_axes_0, beta = encoder_layers_18_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_18_final_layer_norm_weight_to_fp16, x = input_189_cast_fp16)[name = tensor<string, []>("input_191_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_18_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736146432)))]; |
| tensor<fp16, [5120]> encoder_layers_18_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749253696)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_112_cast_fp16 = linear(bias = encoder_layers_18_fc1_bias_to_fp16, weight = encoder_layers_18_fc1_weight_to_fp16, x = input_191_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")]; |
| tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_112_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_18_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749264000)))]; |
| tensor<fp16, [1280]> encoder_layers_18_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762371264)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_113_cast_fp16 = linear(bias = encoder_layers_18_fc2_bias_to_fp16, weight = encoder_layers_18_fc2_weight_to_fp16, x = input_193_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_195_cast_fp16 = add(x = input_189_cast_fp16, y = linear_113_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_119_axes_0 = const()[name = tensor<string, []>("hidden_states_119_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_19_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762373888)))]; |
| tensor<fp16, [1280]> encoder_layers_19_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762376512)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_119_cast_fp16 = layer_norm(axes = hidden_states_119_axes_0, beta = encoder_layers_19_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_19_self_attn_layer_norm_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("hidden_states_119_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762379136)))]; |
| tensor<fp16, [1280]> encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765656000)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_114_cast_fp16 = linear(bias = encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765658624)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768935488)))]; |
| tensor<fp16, [1280]> encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772212352)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_116_cast_fp16 = linear(bias = encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")]; |
| tensor<int32, [4]> var_1140 = const()[name = tensor<string, []>("op_1140"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1141_cast_fp16 = reshape(shape = var_1140, x = linear_114_cast_fp16)[name = tensor<string, []>("op_1141_cast_fp16")]; |
| tensor<int32, [4]> var_1143 = const()[name = tensor<string, []>("op_1143"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1144_cast_fp16 = reshape(shape = var_1143, x = linear_115_cast_fp16)[name = tensor<string, []>("op_1144_cast_fp16")]; |
| tensor<int32, [4]> var_1146 = const()[name = tensor<string, []>("op_1146"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1147_cast_fp16 = reshape(shape = var_1146, x = linear_116_cast_fp16)[name = tensor<string, []>("op_1147_cast_fp16")]; |
| tensor<int32, [4]> value_79_perm_0 = const()[name = tensor<string, []>("value_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_19_cast_fp16 = mul(x = var_1141_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_19_cast_fp16")]; |
| tensor<bool, []> matmul_19_transpose_y_0 = const()[name = tensor<string, []>("matmul_19_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_19_transpose_x_0 = const()[name = tensor<string, []>("matmul_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_166_perm_0 = const()[name = tensor<string, []>("transpose_166_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_167_perm_0 = const()[name = tensor<string, []>("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_167 = transpose(perm = transpose_167_perm_0, x = var_1144_cast_fp16)[name = tensor<string, []>("transpose_241")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_166 = transpose(perm = transpose_166_perm_0, x = mul_19_cast_fp16)[name = tensor<string, []>("transpose_242")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = transpose_166, y = transpose_167)[name = tensor<string, []>("matmul_19_cast_fp16")]; |
| tensor<int32, []> softmax_19_axis_0 = const()[name = tensor<string, []>("softmax_19_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_19_cast_fp16 = softmax(axis = softmax_19_axis_0, x = matmul_19_cast_fp16)[name = tensor<string, []>("softmax_19_cast_fp16")]; |
| tensor<bool, []> attn_output_39_transpose_x_0 = const()[name = tensor<string, []>("attn_output_39_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_39_transpose_y_0 = const()[name = tensor<string, []>("attn_output_39_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_79_cast_fp16 = transpose(perm = value_79_perm_0, x = var_1147_cast_fp16)[name = tensor<string, []>("transpose_243")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_39_cast_fp16 = matmul(transpose_x = attn_output_39_transpose_x_0, transpose_y = attn_output_39_transpose_y_0, x = softmax_19_cast_fp16, y = value_79_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; |
| tensor<int32, [4]> var_1150_perm_0 = const()[name = tensor<string, []>("op_1150_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1152 = const()[name = tensor<string, []>("op_1152"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1150_cast_fp16 = transpose(perm = var_1150_perm_0, x = attn_output_39_cast_fp16)[name = tensor<string, []>("transpose_240")]; |
| tensor<fp16, [1, 1500, 1280]> input_197_cast_fp16 = reshape(shape = var_1152, x = var_1150_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_19_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772214976)))]; |
| tensor<fp16, [1280]> encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775491840)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_117_cast_fp16 = linear(bias = encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_out_proj_weight_to_fp16, x = input_197_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_199_cast_fp16 = add(x = input_195_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")]; |
| tensor<int32, [1]> input_201_axes_0 = const()[name = tensor<string, []>("input_201_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_19_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775494464)))]; |
| tensor<fp16, [1280]> encoder_layers_19_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775497088)))]; |
| tensor<fp16, [1, 1500, 1280]> input_201_cast_fp16 = layer_norm(axes = input_201_axes_0, beta = encoder_layers_19_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_19_final_layer_norm_weight_to_fp16, x = input_199_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_19_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775499712)))]; |
| tensor<fp16, [5120]> encoder_layers_19_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788606976)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_118_cast_fp16 = linear(bias = encoder_layers_19_fc1_bias_to_fp16, weight = encoder_layers_19_fc1_weight_to_fp16, x = input_201_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")]; |
| tensor<string, []> input_203_mode_0 = const()[name = tensor<string, []>("input_203_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_203_cast_fp16 = gelu(mode = input_203_mode_0, x = linear_118_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_19_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788617280)))]; |
| tensor<fp16, [1280]> encoder_layers_19_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801724544)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_119_cast_fp16 = linear(bias = encoder_layers_19_fc2_bias_to_fp16, weight = encoder_layers_19_fc2_weight_to_fp16, x = input_203_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_205_cast_fp16 = add(x = input_199_cast_fp16, y = linear_119_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_125_axes_0 = const()[name = tensor<string, []>("hidden_states_125_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_20_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801727168)))]; |
| tensor<fp16, [1280]> encoder_layers_20_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801729792)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_125_cast_fp16 = layer_norm(axes = hidden_states_125_axes_0, beta = encoder_layers_20_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_20_self_attn_layer_norm_weight_to_fp16, x = input_205_cast_fp16)[name = tensor<string, []>("hidden_states_125_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801732416)))]; |
| tensor<fp16, [1280]> encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805009280)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_120_cast_fp16 = linear(bias = encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805011904)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808288768)))]; |
| tensor<fp16, [1280]> encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811565632)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_122_cast_fp16 = linear(bias = encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_states_125_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")]; |
| tensor<int32, [4]> var_1193 = const()[name = tensor<string, []>("op_1193"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1194_cast_fp16 = reshape(shape = var_1193, x = linear_120_cast_fp16)[name = tensor<string, []>("op_1194_cast_fp16")]; |
| tensor<int32, [4]> var_1196 = const()[name = tensor<string, []>("op_1196"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1197_cast_fp16 = reshape(shape = var_1196, x = linear_121_cast_fp16)[name = tensor<string, []>("op_1197_cast_fp16")]; |
| tensor<int32, [4]> var_1199 = const()[name = tensor<string, []>("op_1199"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1200_cast_fp16 = reshape(shape = var_1199, x = linear_122_cast_fp16)[name = tensor<string, []>("op_1200_cast_fp16")]; |
| tensor<int32, [4]> value_83_perm_0 = const()[name = tensor<string, []>("value_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_20_cast_fp16 = mul(x = var_1194_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_20_cast_fp16")]; |
| tensor<bool, []> matmul_20_transpose_y_0 = const()[name = tensor<string, []>("matmul_20_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_20_transpose_x_0 = const()[name = tensor<string, []>("matmul_20_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_168_perm_0 = const()[name = tensor<string, []>("transpose_168_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_169_perm_0 = const()[name = tensor<string, []>("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_169 = transpose(perm = transpose_169_perm_0, x = var_1197_cast_fp16)[name = tensor<string, []>("transpose_237")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_168 = transpose(perm = transpose_168_perm_0, x = mul_20_cast_fp16)[name = tensor<string, []>("transpose_238")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_168, y = transpose_169)[name = tensor<string, []>("matmul_20_cast_fp16")]; |
| tensor<int32, []> softmax_20_axis_0 = const()[name = tensor<string, []>("softmax_20_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_20_cast_fp16 = softmax(axis = softmax_20_axis_0, x = matmul_20_cast_fp16)[name = tensor<string, []>("softmax_20_cast_fp16")]; |
| tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_83_cast_fp16 = transpose(perm = value_83_perm_0, x = var_1200_cast_fp16)[name = tensor<string, []>("transpose_239")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_20_cast_fp16, y = value_83_cast_fp16)[name = tensor<string, []>("attn_output_41_cast_fp16")]; |
| tensor<int32, [4]> var_1203_perm_0 = const()[name = tensor<string, []>("op_1203_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1205 = const()[name = tensor<string, []>("op_1205"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1203_cast_fp16 = transpose(perm = var_1203_perm_0, x = attn_output_41_cast_fp16)[name = tensor<string, []>("transpose_236")]; |
| tensor<fp16, [1, 1500, 1280]> input_207_cast_fp16 = reshape(shape = var_1205, x = var_1203_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_20_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811568256)))]; |
| tensor<fp16, [1280]> encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814845120)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_123_cast_fp16 = linear(bias = encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_out_proj_weight_to_fp16, x = input_207_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_209_cast_fp16 = add(x = input_205_cast_fp16, y = linear_123_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")]; |
| tensor<int32, [1]> input_211_axes_0 = const()[name = tensor<string, []>("input_211_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_20_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814847744)))]; |
| tensor<fp16, [1280]> encoder_layers_20_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814850368)))]; |
| tensor<fp16, [1, 1500, 1280]> input_211_cast_fp16 = layer_norm(axes = input_211_axes_0, beta = encoder_layers_20_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_20_final_layer_norm_weight_to_fp16, x = input_209_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_20_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814852992)))]; |
| tensor<fp16, [5120]> encoder_layers_20_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827960256)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_124_cast_fp16 = linear(bias = encoder_layers_20_fc1_bias_to_fp16, weight = encoder_layers_20_fc1_weight_to_fp16, x = input_211_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")]; |
| tensor<string, []> input_213_mode_0 = const()[name = tensor<string, []>("input_213_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_213_cast_fp16 = gelu(mode = input_213_mode_0, x = linear_124_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_20_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827970560)))]; |
| tensor<fp16, [1280]> encoder_layers_20_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841077824)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_125_cast_fp16 = linear(bias = encoder_layers_20_fc2_bias_to_fp16, weight = encoder_layers_20_fc2_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_215_cast_fp16 = add(x = input_209_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_131_axes_0 = const()[name = tensor<string, []>("hidden_states_131_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_21_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841080448)))]; |
| tensor<fp16, [1280]> encoder_layers_21_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841083072)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_131_cast_fp16 = layer_norm(axes = hidden_states_131_axes_0, beta = encoder_layers_21_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_21_self_attn_layer_norm_weight_to_fp16, x = input_215_cast_fp16)[name = tensor<string, []>("hidden_states_131_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841085696)))]; |
| tensor<fp16, [1280]> encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844362560)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_126_cast_fp16 = linear(bias = encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_states_131_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844365184)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_states_131_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(847642048)))]; |
| tensor<fp16, [1280]> encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850918912)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_128_cast_fp16 = linear(bias = encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_states_131_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")]; |
| tensor<int32, [4]> var_1246 = const()[name = tensor<string, []>("op_1246"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1247_cast_fp16 = reshape(shape = var_1246, x = linear_126_cast_fp16)[name = tensor<string, []>("op_1247_cast_fp16")]; |
| tensor<int32, [4]> var_1249 = const()[name = tensor<string, []>("op_1249"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1250_cast_fp16 = reshape(shape = var_1249, x = linear_127_cast_fp16)[name = tensor<string, []>("op_1250_cast_fp16")]; |
| tensor<int32, [4]> var_1252 = const()[name = tensor<string, []>("op_1252"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1253_cast_fp16 = reshape(shape = var_1252, x = linear_128_cast_fp16)[name = tensor<string, []>("op_1253_cast_fp16")]; |
| tensor<int32, [4]> value_87_perm_0 = const()[name = tensor<string, []>("value_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_21_cast_fp16 = mul(x = var_1247_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_21_cast_fp16")]; |
| tensor<bool, []> matmul_21_transpose_y_0 = const()[name = tensor<string, []>("matmul_21_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_21_transpose_x_0 = const()[name = tensor<string, []>("matmul_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_170_perm_0 = const()[name = tensor<string, []>("transpose_170_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_171_perm_0 = const()[name = tensor<string, []>("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_171 = transpose(perm = transpose_171_perm_0, x = var_1250_cast_fp16)[name = tensor<string, []>("transpose_233")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_170 = transpose(perm = transpose_170_perm_0, x = mul_21_cast_fp16)[name = tensor<string, []>("transpose_234")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = transpose_170, y = transpose_171)[name = tensor<string, []>("matmul_21_cast_fp16")]; |
| tensor<int32, []> softmax_21_axis_0 = const()[name = tensor<string, []>("softmax_21_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_21_cast_fp16 = softmax(axis = softmax_21_axis_0, x = matmul_21_cast_fp16)[name = tensor<string, []>("softmax_21_cast_fp16")]; |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_87_cast_fp16 = transpose(perm = value_87_perm_0, x = var_1253_cast_fp16)[name = tensor<string, []>("transpose_235")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = softmax_21_cast_fp16, y = value_87_cast_fp16)[name = tensor<string, []>("attn_output_43_cast_fp16")]; |
| tensor<int32, [4]> var_1256_perm_0 = const()[name = tensor<string, []>("op_1256_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1258 = const()[name = tensor<string, []>("op_1258"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1256_cast_fp16 = transpose(perm = var_1256_perm_0, x = attn_output_43_cast_fp16)[name = tensor<string, []>("transpose_232")]; |
| tensor<fp16, [1, 1500, 1280]> input_217_cast_fp16 = reshape(shape = var_1258, x = var_1256_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_21_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850921536)))]; |
| tensor<fp16, [1280]> encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854198400)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_129_cast_fp16 = linear(bias = encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_out_proj_weight_to_fp16, x = input_217_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_219_cast_fp16 = add(x = input_215_cast_fp16, y = linear_129_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")]; |
| tensor<int32, [1]> input_221_axes_0 = const()[name = tensor<string, []>("input_221_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_21_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854201024)))]; |
| tensor<fp16, [1280]> encoder_layers_21_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854203648)))]; |
| tensor<fp16, [1, 1500, 1280]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = encoder_layers_21_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_21_final_layer_norm_weight_to_fp16, x = input_219_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_21_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854206272)))]; |
| tensor<fp16, [5120]> encoder_layers_21_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867313536)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_130_cast_fp16 = linear(bias = encoder_layers_21_fc1_bias_to_fp16, weight = encoder_layers_21_fc1_weight_to_fp16, x = input_221_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")]; |
| tensor<string, []> input_223_mode_0 = const()[name = tensor<string, []>("input_223_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = linear_130_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_21_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867323840)))]; |
| tensor<fp16, [1280]> encoder_layers_21_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880431104)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_131_cast_fp16 = linear(bias = encoder_layers_21_fc2_bias_to_fp16, weight = encoder_layers_21_fc2_weight_to_fp16, x = input_223_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_225_cast_fp16 = add(x = input_219_cast_fp16, y = linear_131_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_137_axes_0 = const()[name = tensor<string, []>("hidden_states_137_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_22_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880433728)))]; |
| tensor<fp16, [1280]> encoder_layers_22_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880436352)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_137_cast_fp16 = layer_norm(axes = hidden_states_137_axes_0, beta = encoder_layers_22_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_22_self_attn_layer_norm_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("hidden_states_137_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880438976)))]; |
| tensor<fp16, [1280]> encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883715840)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_132_cast_fp16 = linear(bias = encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883718464)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886995328)))]; |
| tensor<fp16, [1280]> encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890272192)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_134_cast_fp16 = linear(bias = encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")]; |
| tensor<int32, [4]> var_1299 = const()[name = tensor<string, []>("op_1299"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1300_cast_fp16 = reshape(shape = var_1299, x = linear_132_cast_fp16)[name = tensor<string, []>("op_1300_cast_fp16")]; |
| tensor<int32, [4]> var_1302 = const()[name = tensor<string, []>("op_1302"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1303_cast_fp16 = reshape(shape = var_1302, x = linear_133_cast_fp16)[name = tensor<string, []>("op_1303_cast_fp16")]; |
| tensor<int32, [4]> var_1305 = const()[name = tensor<string, []>("op_1305"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1306_cast_fp16 = reshape(shape = var_1305, x = linear_134_cast_fp16)[name = tensor<string, []>("op_1306_cast_fp16")]; |
| tensor<int32, [4]> value_91_perm_0 = const()[name = tensor<string, []>("value_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_22_cast_fp16 = mul(x = var_1300_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_22_cast_fp16")]; |
| tensor<bool, []> matmul_22_transpose_y_0 = const()[name = tensor<string, []>("matmul_22_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_22_transpose_x_0 = const()[name = tensor<string, []>("matmul_22_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_172_perm_0 = const()[name = tensor<string, []>("transpose_172_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_173_perm_0 = const()[name = tensor<string, []>("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_173 = transpose(perm = transpose_173_perm_0, x = var_1303_cast_fp16)[name = tensor<string, []>("transpose_229")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_172 = transpose(perm = transpose_172_perm_0, x = mul_22_cast_fp16)[name = tensor<string, []>("transpose_230")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_172, y = transpose_173)[name = tensor<string, []>("matmul_22_cast_fp16")]; |
| tensor<int32, []> softmax_22_axis_0 = const()[name = tensor<string, []>("softmax_22_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_22_cast_fp16 = softmax(axis = softmax_22_axis_0, x = matmul_22_cast_fp16)[name = tensor<string, []>("softmax_22_cast_fp16")]; |
| tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_91_cast_fp16 = transpose(perm = value_91_perm_0, x = var_1306_cast_fp16)[name = tensor<string, []>("transpose_231")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_22_cast_fp16, y = value_91_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; |
| tensor<int32, [4]> var_1309_perm_0 = const()[name = tensor<string, []>("op_1309_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1311 = const()[name = tensor<string, []>("op_1311"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1309_cast_fp16 = transpose(perm = var_1309_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_228")]; |
| tensor<fp16, [1, 1500, 1280]> input_227_cast_fp16 = reshape(shape = var_1311, x = var_1309_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_22_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890274816)))]; |
| tensor<fp16, [1280]> encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893551680)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_135_cast_fp16 = linear(bias = encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_out_proj_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_229_cast_fp16 = add(x = input_225_cast_fp16, y = linear_135_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")]; |
| tensor<int32, [1]> input_231_axes_0 = const()[name = tensor<string, []>("input_231_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_22_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893554304)))]; |
| tensor<fp16, [1280]> encoder_layers_22_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893556928)))]; |
| tensor<fp16, [1, 1500, 1280]> input_231_cast_fp16 = layer_norm(axes = input_231_axes_0, beta = encoder_layers_22_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_22_final_layer_norm_weight_to_fp16, x = input_229_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_22_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893559552)))]; |
| tensor<fp16, [5120]> encoder_layers_22_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906666816)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_136_cast_fp16 = linear(bias = encoder_layers_22_fc1_bias_to_fp16, weight = encoder_layers_22_fc1_weight_to_fp16, x = input_231_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")]; |
| tensor<string, []> input_233_mode_0 = const()[name = tensor<string, []>("input_233_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_233_cast_fp16 = gelu(mode = input_233_mode_0, x = linear_136_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_22_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906677120)))]; |
| tensor<fp16, [1280]> encoder_layers_22_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919784384)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_137_cast_fp16 = linear(bias = encoder_layers_22_fc2_bias_to_fp16, weight = encoder_layers_22_fc2_weight_to_fp16, x = input_233_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_235_cast_fp16 = add(x = input_229_cast_fp16, y = linear_137_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_143_axes_0 = const()[name = tensor<string, []>("hidden_states_143_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_23_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919787008)))]; |
| tensor<fp16, [1280]> encoder_layers_23_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919789632)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_143_cast_fp16 = layer_norm(axes = hidden_states_143_axes_0, beta = encoder_layers_23_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_23_self_attn_layer_norm_weight_to_fp16, x = input_235_cast_fp16)[name = tensor<string, []>("hidden_states_143_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919792256)))]; |
| tensor<fp16, [1280]> encoder_layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923069120)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_138_cast_fp16 = linear(bias = encoder_layers_23_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_q_proj_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923071744)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_23_self_attn_k_proj_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926348608)))]; |
| tensor<fp16, [1280]> encoder_layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929625472)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_140_cast_fp16 = linear(bias = encoder_layers_23_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_v_proj_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")]; |
| tensor<int32, [4]> var_1352 = const()[name = tensor<string, []>("op_1352"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1353_cast_fp16 = reshape(shape = var_1352, x = linear_138_cast_fp16)[name = tensor<string, []>("op_1353_cast_fp16")]; |
| tensor<int32, [4]> var_1355 = const()[name = tensor<string, []>("op_1355"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1356_cast_fp16 = reshape(shape = var_1355, x = linear_139_cast_fp16)[name = tensor<string, []>("op_1356_cast_fp16")]; |
| tensor<int32, [4]> var_1358 = const()[name = tensor<string, []>("op_1358"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1359_cast_fp16 = reshape(shape = var_1358, x = linear_140_cast_fp16)[name = tensor<string, []>("op_1359_cast_fp16")]; |
| tensor<int32, [4]> value_95_perm_0 = const()[name = tensor<string, []>("value_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_23_cast_fp16 = mul(x = var_1353_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_23_cast_fp16")]; |
| tensor<bool, []> matmul_23_transpose_y_0 = const()[name = tensor<string, []>("matmul_23_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_23_transpose_x_0 = const()[name = tensor<string, []>("matmul_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_174_perm_0 = const()[name = tensor<string, []>("transpose_174_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_175_perm_0 = const()[name = tensor<string, []>("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_175 = transpose(perm = transpose_175_perm_0, x = var_1356_cast_fp16)[name = tensor<string, []>("transpose_225")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_174 = transpose(perm = transpose_174_perm_0, x = mul_23_cast_fp16)[name = tensor<string, []>("transpose_226")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = transpose_174, y = transpose_175)[name = tensor<string, []>("matmul_23_cast_fp16")]; |
| tensor<int32, []> softmax_23_axis_0 = const()[name = tensor<string, []>("softmax_23_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_23_cast_fp16 = softmax(axis = softmax_23_axis_0, x = matmul_23_cast_fp16)[name = tensor<string, []>("softmax_23_cast_fp16")]; |
| tensor<bool, []> attn_output_47_transpose_x_0 = const()[name = tensor<string, []>("attn_output_47_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_47_transpose_y_0 = const()[name = tensor<string, []>("attn_output_47_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_95_cast_fp16 = transpose(perm = value_95_perm_0, x = var_1359_cast_fp16)[name = tensor<string, []>("transpose_227")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_47_cast_fp16 = matmul(transpose_x = attn_output_47_transpose_x_0, transpose_y = attn_output_47_transpose_y_0, x = softmax_23_cast_fp16, y = value_95_cast_fp16)[name = tensor<string, []>("attn_output_47_cast_fp16")]; |
| tensor<int32, [4]> var_1362_perm_0 = const()[name = tensor<string, []>("op_1362_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1364 = const()[name = tensor<string, []>("op_1364"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1362_cast_fp16 = transpose(perm = var_1362_perm_0, x = attn_output_47_cast_fp16)[name = tensor<string, []>("transpose_224")]; |
| tensor<fp16, [1, 1500, 1280]> input_237_cast_fp16 = reshape(shape = var_1364, x = var_1362_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_23_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929628096)))]; |
| tensor<fp16, [1280]> encoder_layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932904960)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_141_cast_fp16 = linear(bias = encoder_layers_23_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_out_proj_weight_to_fp16, x = input_237_cast_fp16)[name = tensor<string, []>("linear_141_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_239_cast_fp16 = add(x = input_235_cast_fp16, y = linear_141_cast_fp16)[name = tensor<string, []>("input_239_cast_fp16")]; |
| tensor<int32, [1]> input_241_axes_0 = const()[name = tensor<string, []>("input_241_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_23_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932907584)))]; |
| tensor<fp16, [1280]> encoder_layers_23_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932910208)))]; |
| tensor<fp16, [1, 1500, 1280]> input_241_cast_fp16 = layer_norm(axes = input_241_axes_0, beta = encoder_layers_23_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_23_final_layer_norm_weight_to_fp16, x = input_239_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_23_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932912832)))]; |
| tensor<fp16, [5120]> encoder_layers_23_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(946020096)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_142_cast_fp16 = linear(bias = encoder_layers_23_fc1_bias_to_fp16, weight = encoder_layers_23_fc1_weight_to_fp16, x = input_241_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")]; |
| tensor<string, []> input_243_mode_0 = const()[name = tensor<string, []>("input_243_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_243_cast_fp16 = gelu(mode = input_243_mode_0, x = linear_142_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_23_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(946030400)))]; |
| tensor<fp16, [1280]> encoder_layers_23_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959137664)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_143_cast_fp16 = linear(bias = encoder_layers_23_fc2_bias_to_fp16, weight = encoder_layers_23_fc2_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_245_cast_fp16 = add(x = input_239_cast_fp16, y = linear_143_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_149_axes_0 = const()[name = tensor<string, []>("hidden_states_149_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_24_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959140288)))]; |
| tensor<fp16, [1280]> encoder_layers_24_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959142912)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_149_cast_fp16 = layer_norm(axes = hidden_states_149_axes_0, beta = encoder_layers_24_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_24_self_attn_layer_norm_weight_to_fp16, x = input_245_cast_fp16)[name = tensor<string, []>("hidden_states_149_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959145536)))]; |
| tensor<fp16, [1280]> encoder_layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962422400)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_144_cast_fp16 = linear(bias = encoder_layers_24_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_q_proj_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962425024)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_24_self_attn_k_proj_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965701888)))]; |
| tensor<fp16, [1280]> encoder_layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968978752)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_146_cast_fp16 = linear(bias = encoder_layers_24_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_v_proj_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")]; |
| tensor<int32, [4]> var_1405 = const()[name = tensor<string, []>("op_1405"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1406_cast_fp16 = reshape(shape = var_1405, x = linear_144_cast_fp16)[name = tensor<string, []>("op_1406_cast_fp16")]; |
| tensor<int32, [4]> var_1408 = const()[name = tensor<string, []>("op_1408"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1409_cast_fp16 = reshape(shape = var_1408, x = linear_145_cast_fp16)[name = tensor<string, []>("op_1409_cast_fp16")]; |
| tensor<int32, [4]> var_1411 = const()[name = tensor<string, []>("op_1411"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1412_cast_fp16 = reshape(shape = var_1411, x = linear_146_cast_fp16)[name = tensor<string, []>("op_1412_cast_fp16")]; |
| tensor<int32, [4]> value_99_perm_0 = const()[name = tensor<string, []>("value_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_24_cast_fp16 = mul(x = var_1406_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_24_cast_fp16")]; |
| tensor<bool, []> matmul_24_transpose_y_0 = const()[name = tensor<string, []>("matmul_24_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_24_transpose_x_0 = const()[name = tensor<string, []>("matmul_24_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_176_perm_0 = const()[name = tensor<string, []>("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_177_perm_0 = const()[name = tensor<string, []>("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_177 = transpose(perm = transpose_177_perm_0, x = var_1409_cast_fp16)[name = tensor<string, []>("transpose_221")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_176 = transpose(perm = transpose_176_perm_0, x = mul_24_cast_fp16)[name = tensor<string, []>("transpose_222")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_24_cast_fp16 = matmul(transpose_x = matmul_24_transpose_x_0, transpose_y = matmul_24_transpose_y_0, x = transpose_176, y = transpose_177)[name = tensor<string, []>("matmul_24_cast_fp16")]; |
| tensor<int32, []> softmax_24_axis_0 = const()[name = tensor<string, []>("softmax_24_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_24_cast_fp16 = softmax(axis = softmax_24_axis_0, x = matmul_24_cast_fp16)[name = tensor<string, []>("softmax_24_cast_fp16")]; |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_99_cast_fp16 = transpose(perm = value_99_perm_0, x = var_1412_cast_fp16)[name = tensor<string, []>("transpose_223")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = softmax_24_cast_fp16, y = value_99_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")]; |
| tensor<int32, [4]> var_1415_perm_0 = const()[name = tensor<string, []>("op_1415_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1417 = const()[name = tensor<string, []>("op_1417"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1415_cast_fp16 = transpose(perm = var_1415_perm_0, x = attn_output_49_cast_fp16)[name = tensor<string, []>("transpose_220")]; |
| tensor<fp16, [1, 1500, 1280]> input_247_cast_fp16 = reshape(shape = var_1417, x = var_1415_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_24_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968981376)))]; |
| tensor<fp16, [1280]> encoder_layers_24_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972258240)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_147_cast_fp16 = linear(bias = encoder_layers_24_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_out_proj_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_249_cast_fp16 = add(x = input_245_cast_fp16, y = linear_147_cast_fp16)[name = tensor<string, []>("input_249_cast_fp16")]; |
| tensor<int32, [1]> input_251_axes_0 = const()[name = tensor<string, []>("input_251_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_24_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972260864)))]; |
| tensor<fp16, [1280]> encoder_layers_24_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972263488)))]; |
| tensor<fp16, [1, 1500, 1280]> input_251_cast_fp16 = layer_norm(axes = input_251_axes_0, beta = encoder_layers_24_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_24_final_layer_norm_weight_to_fp16, x = input_249_cast_fp16)[name = tensor<string, []>("input_251_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_24_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972266112)))]; |
| tensor<fp16, [5120]> encoder_layers_24_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985373376)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_148_cast_fp16 = linear(bias = encoder_layers_24_fc1_bias_to_fp16, weight = encoder_layers_24_fc1_weight_to_fp16, x = input_251_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")]; |
| tensor<string, []> input_253_mode_0 = const()[name = tensor<string, []>("input_253_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_253_cast_fp16 = gelu(mode = input_253_mode_0, x = linear_148_cast_fp16)[name = tensor<string, []>("input_253_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_24_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985383680)))]; |
| tensor<fp16, [1280]> encoder_layers_24_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998490944)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_149_cast_fp16 = linear(bias = encoder_layers_24_fc2_bias_to_fp16, weight = encoder_layers_24_fc2_weight_to_fp16, x = input_253_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_255_cast_fp16 = add(x = input_249_cast_fp16, y = linear_149_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_155_axes_0 = const()[name = tensor<string, []>("hidden_states_155_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_25_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998493568)))]; |
| tensor<fp16, [1280]> encoder_layers_25_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998496192)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_155_cast_fp16 = layer_norm(axes = hidden_states_155_axes_0, beta = encoder_layers_25_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_25_self_attn_layer_norm_weight_to_fp16, x = input_255_cast_fp16)[name = tensor<string, []>("hidden_states_155_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998498816)))]; |
| tensor<fp16, [1280]> encoder_layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001775680)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_150_cast_fp16 = linear(bias = encoder_layers_25_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_q_proj_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor<string, []>("linear_150_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001778304)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_25_self_attn_k_proj_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005055168)))]; |
| tensor<fp16, [1280]> encoder_layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008332032)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_152_cast_fp16 = linear(bias = encoder_layers_25_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_v_proj_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")]; |
| tensor<int32, [4]> var_1458 = const()[name = tensor<string, []>("op_1458"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1459_cast_fp16 = reshape(shape = var_1458, x = linear_150_cast_fp16)[name = tensor<string, []>("op_1459_cast_fp16")]; |
| tensor<int32, [4]> var_1461 = const()[name = tensor<string, []>("op_1461"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1462_cast_fp16 = reshape(shape = var_1461, x = linear_151_cast_fp16)[name = tensor<string, []>("op_1462_cast_fp16")]; |
| tensor<int32, [4]> var_1464 = const()[name = tensor<string, []>("op_1464"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1465_cast_fp16 = reshape(shape = var_1464, x = linear_152_cast_fp16)[name = tensor<string, []>("op_1465_cast_fp16")]; |
| tensor<int32, [4]> value_103_perm_0 = const()[name = tensor<string, []>("value_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_25_cast_fp16 = mul(x = var_1459_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_25_cast_fp16")]; |
| tensor<bool, []> matmul_25_transpose_y_0 = const()[name = tensor<string, []>("matmul_25_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_25_transpose_x_0 = const()[name = tensor<string, []>("matmul_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_178_perm_0 = const()[name = tensor<string, []>("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_179_perm_0 = const()[name = tensor<string, []>("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_179 = transpose(perm = transpose_179_perm_0, x = var_1462_cast_fp16)[name = tensor<string, []>("transpose_217")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_178 = transpose(perm = transpose_178_perm_0, x = mul_25_cast_fp16)[name = tensor<string, []>("transpose_218")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_25_cast_fp16 = matmul(transpose_x = matmul_25_transpose_x_0, transpose_y = matmul_25_transpose_y_0, x = transpose_178, y = transpose_179)[name = tensor<string, []>("matmul_25_cast_fp16")]; |
| tensor<int32, []> softmax_25_axis_0 = const()[name = tensor<string, []>("softmax_25_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_25_cast_fp16 = softmax(axis = softmax_25_axis_0, x = matmul_25_cast_fp16)[name = tensor<string, []>("softmax_25_cast_fp16")]; |
| tensor<bool, []> attn_output_51_transpose_x_0 = const()[name = tensor<string, []>("attn_output_51_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_51_transpose_y_0 = const()[name = tensor<string, []>("attn_output_51_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_103_cast_fp16 = transpose(perm = value_103_perm_0, x = var_1465_cast_fp16)[name = tensor<string, []>("transpose_219")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_51_cast_fp16 = matmul(transpose_x = attn_output_51_transpose_x_0, transpose_y = attn_output_51_transpose_y_0, x = softmax_25_cast_fp16, y = value_103_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; |
| tensor<int32, [4]> var_1468_perm_0 = const()[name = tensor<string, []>("op_1468_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1470 = const()[name = tensor<string, []>("op_1470"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1468_cast_fp16 = transpose(perm = var_1468_perm_0, x = attn_output_51_cast_fp16)[name = tensor<string, []>("transpose_216")]; |
| tensor<fp16, [1, 1500, 1280]> input_257_cast_fp16 = reshape(shape = var_1470, x = var_1468_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_25_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008334656)))]; |
| tensor<fp16, [1280]> encoder_layers_25_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011611520)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_153_cast_fp16 = linear(bias = encoder_layers_25_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_out_proj_weight_to_fp16, x = input_257_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_259_cast_fp16 = add(x = input_255_cast_fp16, y = linear_153_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")]; |
| tensor<int32, [1]> input_261_axes_0 = const()[name = tensor<string, []>("input_261_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_25_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011614144)))]; |
| tensor<fp16, [1280]> encoder_layers_25_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011616768)))]; |
| tensor<fp16, [1, 1500, 1280]> input_261_cast_fp16 = layer_norm(axes = input_261_axes_0, beta = encoder_layers_25_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_25_final_layer_norm_weight_to_fp16, x = input_259_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_25_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011619392)))]; |
| tensor<fp16, [5120]> encoder_layers_25_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024726656)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_154_cast_fp16 = linear(bias = encoder_layers_25_fc1_bias_to_fp16, weight = encoder_layers_25_fc1_weight_to_fp16, x = input_261_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")]; |
| tensor<string, []> input_263_mode_0 = const()[name = tensor<string, []>("input_263_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_263_cast_fp16 = gelu(mode = input_263_mode_0, x = linear_154_cast_fp16)[name = tensor<string, []>("input_263_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_25_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024736960)))]; |
| tensor<fp16, [1280]> encoder_layers_25_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037844224)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_155_cast_fp16 = linear(bias = encoder_layers_25_fc2_bias_to_fp16, weight = encoder_layers_25_fc2_weight_to_fp16, x = input_263_cast_fp16)[name = tensor<string, []>("linear_155_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_265_cast_fp16 = add(x = input_259_cast_fp16, y = linear_155_cast_fp16)[name = tensor<string, []>("input_265_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_161_axes_0 = const()[name = tensor<string, []>("hidden_states_161_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_26_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037846848)))]; |
| tensor<fp16, [1280]> encoder_layers_26_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037849472)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_161_cast_fp16 = layer_norm(axes = hidden_states_161_axes_0, beta = encoder_layers_26_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_26_self_attn_layer_norm_weight_to_fp16, x = input_265_cast_fp16)[name = tensor<string, []>("hidden_states_161_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037852096)))]; |
| tensor<fp16, [1280]> encoder_layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041128960)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_156_cast_fp16 = linear(bias = encoder_layers_26_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_q_proj_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041131584)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_26_self_attn_k_proj_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044408448)))]; |
| tensor<fp16, [1280]> encoder_layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047685312)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_158_cast_fp16 = linear(bias = encoder_layers_26_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_v_proj_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")]; |
| tensor<int32, [4]> var_1511 = const()[name = tensor<string, []>("op_1511"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1512_cast_fp16 = reshape(shape = var_1511, x = linear_156_cast_fp16)[name = tensor<string, []>("op_1512_cast_fp16")]; |
| tensor<int32, [4]> var_1514 = const()[name = tensor<string, []>("op_1514"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1515_cast_fp16 = reshape(shape = var_1514, x = linear_157_cast_fp16)[name = tensor<string, []>("op_1515_cast_fp16")]; |
| tensor<int32, [4]> var_1517 = const()[name = tensor<string, []>("op_1517"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1518_cast_fp16 = reshape(shape = var_1517, x = linear_158_cast_fp16)[name = tensor<string, []>("op_1518_cast_fp16")]; |
| tensor<int32, [4]> value_107_perm_0 = const()[name = tensor<string, []>("value_107_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_26_cast_fp16 = mul(x = var_1512_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_26_cast_fp16")]; |
| tensor<bool, []> matmul_26_transpose_y_0 = const()[name = tensor<string, []>("matmul_26_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_26_transpose_x_0 = const()[name = tensor<string, []>("matmul_26_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_180_perm_0 = const()[name = tensor<string, []>("transpose_180_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_181_perm_0 = const()[name = tensor<string, []>("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_181 = transpose(perm = transpose_181_perm_0, x = var_1515_cast_fp16)[name = tensor<string, []>("transpose_213")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_180 = transpose(perm = transpose_180_perm_0, x = mul_26_cast_fp16)[name = tensor<string, []>("transpose_214")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_26_cast_fp16 = matmul(transpose_x = matmul_26_transpose_x_0, transpose_y = matmul_26_transpose_y_0, x = transpose_180, y = transpose_181)[name = tensor<string, []>("matmul_26_cast_fp16")]; |
| tensor<int32, []> softmax_26_axis_0 = const()[name = tensor<string, []>("softmax_26_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_26_cast_fp16 = softmax(axis = softmax_26_axis_0, x = matmul_26_cast_fp16)[name = tensor<string, []>("softmax_26_cast_fp16")]; |
| tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_107_cast_fp16 = transpose(perm = value_107_perm_0, x = var_1518_cast_fp16)[name = tensor<string, []>("transpose_215")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_53_cast_fp16 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = softmax_26_cast_fp16, y = value_107_cast_fp16)[name = tensor<string, []>("attn_output_53_cast_fp16")]; |
| tensor<int32, [4]> var_1521_perm_0 = const()[name = tensor<string, []>("op_1521_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1523 = const()[name = tensor<string, []>("op_1523"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1521_cast_fp16 = transpose(perm = var_1521_perm_0, x = attn_output_53_cast_fp16)[name = tensor<string, []>("transpose_212")]; |
| tensor<fp16, [1, 1500, 1280]> input_267_cast_fp16 = reshape(shape = var_1523, x = var_1521_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_26_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047687936)))]; |
| tensor<fp16, [1280]> encoder_layers_26_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050964800)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_159_cast_fp16 = linear(bias = encoder_layers_26_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_out_proj_weight_to_fp16, x = input_267_cast_fp16)[name = tensor<string, []>("linear_159_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_269_cast_fp16 = add(x = input_265_cast_fp16, y = linear_159_cast_fp16)[name = tensor<string, []>("input_269_cast_fp16")]; |
| tensor<int32, [1]> input_271_axes_0 = const()[name = tensor<string, []>("input_271_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_26_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050967424)))]; |
| tensor<fp16, [1280]> encoder_layers_26_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050970048)))]; |
| tensor<fp16, [1, 1500, 1280]> input_271_cast_fp16 = layer_norm(axes = input_271_axes_0, beta = encoder_layers_26_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_26_final_layer_norm_weight_to_fp16, x = input_269_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_26_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050972672)))]; |
| tensor<fp16, [5120]> encoder_layers_26_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064079936)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_160_cast_fp16 = linear(bias = encoder_layers_26_fc1_bias_to_fp16, weight = encoder_layers_26_fc1_weight_to_fp16, x = input_271_cast_fp16)[name = tensor<string, []>("linear_160_cast_fp16")]; |
| tensor<string, []> input_273_mode_0 = const()[name = tensor<string, []>("input_273_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_273_cast_fp16 = gelu(mode = input_273_mode_0, x = linear_160_cast_fp16)[name = tensor<string, []>("input_273_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_26_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064090240)))]; |
| tensor<fp16, [1280]> encoder_layers_26_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077197504)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_161_cast_fp16 = linear(bias = encoder_layers_26_fc2_bias_to_fp16, weight = encoder_layers_26_fc2_weight_to_fp16, x = input_273_cast_fp16)[name = tensor<string, []>("linear_161_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_275_cast_fp16 = add(x = input_269_cast_fp16, y = linear_161_cast_fp16)[name = tensor<string, []>("input_275_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_167_axes_0 = const()[name = tensor<string, []>("hidden_states_167_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_27_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077200128)))]; |
| tensor<fp16, [1280]> encoder_layers_27_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077202752)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_167_cast_fp16 = layer_norm(axes = hidden_states_167_axes_0, beta = encoder_layers_27_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_27_self_attn_layer_norm_weight_to_fp16, x = input_275_cast_fp16)[name = tensor<string, []>("hidden_states_167_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077205376)))]; |
| tensor<fp16, [1280]> encoder_layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080482240)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_162_cast_fp16 = linear(bias = encoder_layers_27_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_27_self_attn_q_proj_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor<string, []>("linear_162_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080484864)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_27_self_attn_k_proj_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor<string, []>("linear_163_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1083761728)))]; |
| tensor<fp16, [1280]> encoder_layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087038592)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_164_cast_fp16 = linear(bias = encoder_layers_27_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_27_self_attn_v_proj_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor<string, []>("linear_164_cast_fp16")]; |
| tensor<int32, [4]> var_1564 = const()[name = tensor<string, []>("op_1564"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1565_cast_fp16 = reshape(shape = var_1564, x = linear_162_cast_fp16)[name = tensor<string, []>("op_1565_cast_fp16")]; |
| tensor<int32, [4]> var_1567 = const()[name = tensor<string, []>("op_1567"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1568_cast_fp16 = reshape(shape = var_1567, x = linear_163_cast_fp16)[name = tensor<string, []>("op_1568_cast_fp16")]; |
| tensor<int32, [4]> var_1570 = const()[name = tensor<string, []>("op_1570"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1571_cast_fp16 = reshape(shape = var_1570, x = linear_164_cast_fp16)[name = tensor<string, []>("op_1571_cast_fp16")]; |
| tensor<int32, [4]> value_111_perm_0 = const()[name = tensor<string, []>("value_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_27_cast_fp16 = mul(x = var_1565_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_27_cast_fp16")]; |
| tensor<bool, []> matmul_27_transpose_y_0 = const()[name = tensor<string, []>("matmul_27_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_27_transpose_x_0 = const()[name = tensor<string, []>("matmul_27_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_182_perm_0 = const()[name = tensor<string, []>("transpose_182_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_183_perm_0 = const()[name = tensor<string, []>("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_183 = transpose(perm = transpose_183_perm_0, x = var_1568_cast_fp16)[name = tensor<string, []>("transpose_209")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_182 = transpose(perm = transpose_182_perm_0, x = mul_27_cast_fp16)[name = tensor<string, []>("transpose_210")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_27_cast_fp16 = matmul(transpose_x = matmul_27_transpose_x_0, transpose_y = matmul_27_transpose_y_0, x = transpose_182, y = transpose_183)[name = tensor<string, []>("matmul_27_cast_fp16")]; |
| tensor<int32, []> softmax_27_axis_0 = const()[name = tensor<string, []>("softmax_27_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_27_cast_fp16 = softmax(axis = softmax_27_axis_0, x = matmul_27_cast_fp16)[name = tensor<string, []>("softmax_27_cast_fp16")]; |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_111_cast_fp16 = transpose(perm = value_111_perm_0, x = var_1571_cast_fp16)[name = tensor<string, []>("transpose_211")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = softmax_27_cast_fp16, y = value_111_cast_fp16)[name = tensor<string, []>("attn_output_55_cast_fp16")]; |
| tensor<int32, [4]> var_1574_perm_0 = const()[name = tensor<string, []>("op_1574_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1576 = const()[name = tensor<string, []>("op_1576"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1574_cast_fp16 = transpose(perm = var_1574_perm_0, x = attn_output_55_cast_fp16)[name = tensor<string, []>("transpose_208")]; |
| tensor<fp16, [1, 1500, 1280]> input_277_cast_fp16 = reshape(shape = var_1576, x = var_1574_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_27_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087041216)))]; |
| tensor<fp16, [1280]> encoder_layers_27_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090318080)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_165_cast_fp16 = linear(bias = encoder_layers_27_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_27_self_attn_out_proj_weight_to_fp16, x = input_277_cast_fp16)[name = tensor<string, []>("linear_165_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_279_cast_fp16 = add(x = input_275_cast_fp16, y = linear_165_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")]; |
| tensor<int32, [1]> input_281_axes_0 = const()[name = tensor<string, []>("input_281_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_27_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090320704)))]; |
| tensor<fp16, [1280]> encoder_layers_27_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090323328)))]; |
| tensor<fp16, [1, 1500, 1280]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = encoder_layers_27_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_27_final_layer_norm_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("input_281_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_27_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090325952)))]; |
| tensor<fp16, [5120]> encoder_layers_27_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103433216)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_166_cast_fp16 = linear(bias = encoder_layers_27_fc1_bias_to_fp16, weight = encoder_layers_27_fc1_weight_to_fp16, x = input_281_cast_fp16)[name = tensor<string, []>("linear_166_cast_fp16")]; |
| tensor<string, []> input_283_mode_0 = const()[name = tensor<string, []>("input_283_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_283_cast_fp16 = gelu(mode = input_283_mode_0, x = linear_166_cast_fp16)[name = tensor<string, []>("input_283_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_27_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103443520)))]; |
| tensor<fp16, [1280]> encoder_layers_27_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_27_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116550784)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_167_cast_fp16 = linear(bias = encoder_layers_27_fc2_bias_to_fp16, weight = encoder_layers_27_fc2_weight_to_fp16, x = input_283_cast_fp16)[name = tensor<string, []>("linear_167_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_285_cast_fp16 = add(x = input_279_cast_fp16, y = linear_167_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_173_axes_0 = const()[name = tensor<string, []>("hidden_states_173_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_28_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116553408)))]; |
| tensor<fp16, [1280]> encoder_layers_28_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116556032)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_173_cast_fp16 = layer_norm(axes = hidden_states_173_axes_0, beta = encoder_layers_28_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_28_self_attn_layer_norm_weight_to_fp16, x = input_285_cast_fp16)[name = tensor<string, []>("hidden_states_173_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116558656)))]; |
| tensor<fp16, [1280]> encoder_layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119835520)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_168_cast_fp16 = linear(bias = encoder_layers_28_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_28_self_attn_q_proj_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor<string, []>("linear_168_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119838144)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_28_self_attn_k_proj_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor<string, []>("linear_169_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123115008)))]; |
| tensor<fp16, [1280]> encoder_layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126391872)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_170_cast_fp16 = linear(bias = encoder_layers_28_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_28_self_attn_v_proj_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor<string, []>("linear_170_cast_fp16")]; |
| tensor<int32, [4]> var_1617 = const()[name = tensor<string, []>("op_1617"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1618_cast_fp16 = reshape(shape = var_1617, x = linear_168_cast_fp16)[name = tensor<string, []>("op_1618_cast_fp16")]; |
| tensor<int32, [4]> var_1620 = const()[name = tensor<string, []>("op_1620"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1621_cast_fp16 = reshape(shape = var_1620, x = linear_169_cast_fp16)[name = tensor<string, []>("op_1621_cast_fp16")]; |
| tensor<int32, [4]> var_1623 = const()[name = tensor<string, []>("op_1623"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1624_cast_fp16 = reshape(shape = var_1623, x = linear_170_cast_fp16)[name = tensor<string, []>("op_1624_cast_fp16")]; |
| tensor<int32, [4]> value_115_perm_0 = const()[name = tensor<string, []>("value_115_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_28_cast_fp16 = mul(x = var_1618_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_28_cast_fp16")]; |
| tensor<bool, []> matmul_28_transpose_y_0 = const()[name = tensor<string, []>("matmul_28_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_28_transpose_x_0 = const()[name = tensor<string, []>("matmul_28_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_184_perm_0 = const()[name = tensor<string, []>("transpose_184_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_185_perm_0 = const()[name = tensor<string, []>("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_185 = transpose(perm = transpose_185_perm_0, x = var_1621_cast_fp16)[name = tensor<string, []>("transpose_205")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_184 = transpose(perm = transpose_184_perm_0, x = mul_28_cast_fp16)[name = tensor<string, []>("transpose_206")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_28_cast_fp16 = matmul(transpose_x = matmul_28_transpose_x_0, transpose_y = matmul_28_transpose_y_0, x = transpose_184, y = transpose_185)[name = tensor<string, []>("matmul_28_cast_fp16")]; |
| tensor<int32, []> softmax_28_axis_0 = const()[name = tensor<string, []>("softmax_28_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_28_cast_fp16 = softmax(axis = softmax_28_axis_0, x = matmul_28_cast_fp16)[name = tensor<string, []>("softmax_28_cast_fp16")]; |
| tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_115_cast_fp16 = transpose(perm = value_115_perm_0, x = var_1624_cast_fp16)[name = tensor<string, []>("transpose_207")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_57_cast_fp16 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = softmax_28_cast_fp16, y = value_115_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; |
| tensor<int32, [4]> var_1627_perm_0 = const()[name = tensor<string, []>("op_1627_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1629 = const()[name = tensor<string, []>("op_1629"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1627_cast_fp16 = transpose(perm = var_1627_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_204")]; |
| tensor<fp16, [1, 1500, 1280]> input_287_cast_fp16 = reshape(shape = var_1629, x = var_1627_cast_fp16)[name = tensor<string, []>("input_287_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_28_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126394496)))]; |
| tensor<fp16, [1280]> encoder_layers_28_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129671360)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_171_cast_fp16 = linear(bias = encoder_layers_28_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_28_self_attn_out_proj_weight_to_fp16, x = input_287_cast_fp16)[name = tensor<string, []>("linear_171_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_289_cast_fp16 = add(x = input_285_cast_fp16, y = linear_171_cast_fp16)[name = tensor<string, []>("input_289_cast_fp16")]; |
| tensor<int32, [1]> input_291_axes_0 = const()[name = tensor<string, []>("input_291_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_28_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129673984)))]; |
| tensor<fp16, [1280]> encoder_layers_28_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129676608)))]; |
| tensor<fp16, [1, 1500, 1280]> input_291_cast_fp16 = layer_norm(axes = input_291_axes_0, beta = encoder_layers_28_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_28_final_layer_norm_weight_to_fp16, x = input_289_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_28_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129679232)))]; |
| tensor<fp16, [5120]> encoder_layers_28_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142786496)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_172_cast_fp16 = linear(bias = encoder_layers_28_fc1_bias_to_fp16, weight = encoder_layers_28_fc1_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("linear_172_cast_fp16")]; |
| tensor<string, []> input_293_mode_0 = const()[name = tensor<string, []>("input_293_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_293_cast_fp16 = gelu(mode = input_293_mode_0, x = linear_172_cast_fp16)[name = tensor<string, []>("input_293_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_28_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142796800)))]; |
| tensor<fp16, [1280]> encoder_layers_28_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_28_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155904064)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_173_cast_fp16 = linear(bias = encoder_layers_28_fc2_bias_to_fp16, weight = encoder_layers_28_fc2_weight_to_fp16, x = input_293_cast_fp16)[name = tensor<string, []>("linear_173_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_295_cast_fp16 = add(x = input_289_cast_fp16, y = linear_173_cast_fp16)[name = tensor<string, []>("input_295_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_179_axes_0 = const()[name = tensor<string, []>("hidden_states_179_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_29_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155906688)))]; |
| tensor<fp16, [1280]> encoder_layers_29_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155909312)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_179_cast_fp16 = layer_norm(axes = hidden_states_179_axes_0, beta = encoder_layers_29_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_29_self_attn_layer_norm_weight_to_fp16, x = input_295_cast_fp16)[name = tensor<string, []>("hidden_states_179_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155911936)))]; |
| tensor<fp16, [1280]> encoder_layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159188800)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_174_cast_fp16 = linear(bias = encoder_layers_29_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_29_self_attn_q_proj_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor<string, []>("linear_174_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159191424)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_29_self_attn_k_proj_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor<string, []>("linear_175_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162468288)))]; |
| tensor<fp16, [1280]> encoder_layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165745152)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_176_cast_fp16 = linear(bias = encoder_layers_29_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_29_self_attn_v_proj_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor<string, []>("linear_176_cast_fp16")]; |
| tensor<int32, [4]> var_1670 = const()[name = tensor<string, []>("op_1670"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1671_cast_fp16 = reshape(shape = var_1670, x = linear_174_cast_fp16)[name = tensor<string, []>("op_1671_cast_fp16")]; |
| tensor<int32, [4]> var_1673 = const()[name = tensor<string, []>("op_1673"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1674_cast_fp16 = reshape(shape = var_1673, x = linear_175_cast_fp16)[name = tensor<string, []>("op_1674_cast_fp16")]; |
| tensor<int32, [4]> var_1676 = const()[name = tensor<string, []>("op_1676"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1677_cast_fp16 = reshape(shape = var_1676, x = linear_176_cast_fp16)[name = tensor<string, []>("op_1677_cast_fp16")]; |
| tensor<int32, [4]> value_119_perm_0 = const()[name = tensor<string, []>("value_119_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_29_cast_fp16 = mul(x = var_1671_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_29_cast_fp16")]; |
| tensor<bool, []> matmul_29_transpose_y_0 = const()[name = tensor<string, []>("matmul_29_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_29_transpose_x_0 = const()[name = tensor<string, []>("matmul_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_186_perm_0 = const()[name = tensor<string, []>("transpose_186_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_187_perm_0 = const()[name = tensor<string, []>("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_187 = transpose(perm = transpose_187_perm_0, x = var_1674_cast_fp16)[name = tensor<string, []>("transpose_201")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_186 = transpose(perm = transpose_186_perm_0, x = mul_29_cast_fp16)[name = tensor<string, []>("transpose_202")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_29_cast_fp16 = matmul(transpose_x = matmul_29_transpose_x_0, transpose_y = matmul_29_transpose_y_0, x = transpose_186, y = transpose_187)[name = tensor<string, []>("matmul_29_cast_fp16")]; |
| tensor<int32, []> softmax_29_axis_0 = const()[name = tensor<string, []>("softmax_29_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_29_cast_fp16 = softmax(axis = softmax_29_axis_0, x = matmul_29_cast_fp16)[name = tensor<string, []>("softmax_29_cast_fp16")]; |
| tensor<bool, []> attn_output_59_transpose_x_0 = const()[name = tensor<string, []>("attn_output_59_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_59_transpose_y_0 = const()[name = tensor<string, []>("attn_output_59_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_119_cast_fp16 = transpose(perm = value_119_perm_0, x = var_1677_cast_fp16)[name = tensor<string, []>("transpose_203")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_59_cast_fp16 = matmul(transpose_x = attn_output_59_transpose_x_0, transpose_y = attn_output_59_transpose_y_0, x = softmax_29_cast_fp16, y = value_119_cast_fp16)[name = tensor<string, []>("attn_output_59_cast_fp16")]; |
| tensor<int32, [4]> var_1680_perm_0 = const()[name = tensor<string, []>("op_1680_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1682 = const()[name = tensor<string, []>("op_1682"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1680_cast_fp16 = transpose(perm = var_1680_perm_0, x = attn_output_59_cast_fp16)[name = tensor<string, []>("transpose_200")]; |
| tensor<fp16, [1, 1500, 1280]> input_297_cast_fp16 = reshape(shape = var_1682, x = var_1680_cast_fp16)[name = tensor<string, []>("input_297_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_29_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165747776)))]; |
| tensor<fp16, [1280]> encoder_layers_29_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169024640)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_177_cast_fp16 = linear(bias = encoder_layers_29_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_29_self_attn_out_proj_weight_to_fp16, x = input_297_cast_fp16)[name = tensor<string, []>("linear_177_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_299_cast_fp16 = add(x = input_295_cast_fp16, y = linear_177_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")]; |
| tensor<int32, [1]> input_301_axes_0 = const()[name = tensor<string, []>("input_301_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_29_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169027264)))]; |
| tensor<fp16, [1280]> encoder_layers_29_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169029888)))]; |
| tensor<fp16, [1, 1500, 1280]> input_301_cast_fp16 = layer_norm(axes = input_301_axes_0, beta = encoder_layers_29_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_29_final_layer_norm_weight_to_fp16, x = input_299_cast_fp16)[name = tensor<string, []>("input_301_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_29_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169032512)))]; |
| tensor<fp16, [5120]> encoder_layers_29_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1182139776)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_178_cast_fp16 = linear(bias = encoder_layers_29_fc1_bias_to_fp16, weight = encoder_layers_29_fc1_weight_to_fp16, x = input_301_cast_fp16)[name = tensor<string, []>("linear_178_cast_fp16")]; |
| tensor<string, []> input_303_mode_0 = const()[name = tensor<string, []>("input_303_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_303_cast_fp16 = gelu(mode = input_303_mode_0, x = linear_178_cast_fp16)[name = tensor<string, []>("input_303_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_29_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1182150080)))]; |
| tensor<fp16, [1280]> encoder_layers_29_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_29_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195257344)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_179_cast_fp16 = linear(bias = encoder_layers_29_fc2_bias_to_fp16, weight = encoder_layers_29_fc2_weight_to_fp16, x = input_303_cast_fp16)[name = tensor<string, []>("linear_179_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_305_cast_fp16 = add(x = input_299_cast_fp16, y = linear_179_cast_fp16)[name = tensor<string, []>("input_305_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_185_axes_0 = const()[name = tensor<string, []>("hidden_states_185_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_30_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195259968)))]; |
| tensor<fp16, [1280]> encoder_layers_30_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195262592)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_185_cast_fp16 = layer_norm(axes = hidden_states_185_axes_0, beta = encoder_layers_30_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_30_self_attn_layer_norm_weight_to_fp16, x = input_305_cast_fp16)[name = tensor<string, []>("hidden_states_185_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195265216)))]; |
| tensor<fp16, [1280]> encoder_layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198542080)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_180_cast_fp16 = linear(bias = encoder_layers_30_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_30_self_attn_q_proj_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor<string, []>("linear_180_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198544704)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_30_self_attn_k_proj_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor<string, []>("linear_181_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1201821568)))]; |
| tensor<fp16, [1280]> encoder_layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205098432)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_182_cast_fp16 = linear(bias = encoder_layers_30_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_30_self_attn_v_proj_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor<string, []>("linear_182_cast_fp16")]; |
| tensor<int32, [4]> var_1723 = const()[name = tensor<string, []>("op_1723"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1724_cast_fp16 = reshape(shape = var_1723, x = linear_180_cast_fp16)[name = tensor<string, []>("op_1724_cast_fp16")]; |
| tensor<int32, [4]> var_1726 = const()[name = tensor<string, []>("op_1726"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1727_cast_fp16 = reshape(shape = var_1726, x = linear_181_cast_fp16)[name = tensor<string, []>("op_1727_cast_fp16")]; |
| tensor<int32, [4]> var_1729 = const()[name = tensor<string, []>("op_1729"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1730_cast_fp16 = reshape(shape = var_1729, x = linear_182_cast_fp16)[name = tensor<string, []>("op_1730_cast_fp16")]; |
| tensor<int32, [4]> value_123_perm_0 = const()[name = tensor<string, []>("value_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_30_cast_fp16 = mul(x = var_1724_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_30_cast_fp16")]; |
| tensor<bool, []> matmul_30_transpose_y_0 = const()[name = tensor<string, []>("matmul_30_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_30_transpose_x_0 = const()[name = tensor<string, []>("matmul_30_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_188_perm_0 = const()[name = tensor<string, []>("transpose_188_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_189_perm_0 = const()[name = tensor<string, []>("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_189 = transpose(perm = transpose_189_perm_0, x = var_1727_cast_fp16)[name = tensor<string, []>("transpose_197")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_188 = transpose(perm = transpose_188_perm_0, x = mul_30_cast_fp16)[name = tensor<string, []>("transpose_198")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_30_cast_fp16 = matmul(transpose_x = matmul_30_transpose_x_0, transpose_y = matmul_30_transpose_y_0, x = transpose_188, y = transpose_189)[name = tensor<string, []>("matmul_30_cast_fp16")]; |
| tensor<int32, []> softmax_30_axis_0 = const()[name = tensor<string, []>("softmax_30_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_30_cast_fp16 = softmax(axis = softmax_30_axis_0, x = matmul_30_cast_fp16)[name = tensor<string, []>("softmax_30_cast_fp16")]; |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_123_cast_fp16 = transpose(perm = value_123_perm_0, x = var_1730_cast_fp16)[name = tensor<string, []>("transpose_199")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = softmax_30_cast_fp16, y = value_123_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")]; |
| tensor<int32, [4]> var_1733_perm_0 = const()[name = tensor<string, []>("op_1733_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1735 = const()[name = tensor<string, []>("op_1735"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1733_cast_fp16 = transpose(perm = var_1733_perm_0, x = attn_output_61_cast_fp16)[name = tensor<string, []>("transpose_196")]; |
| tensor<fp16, [1, 1500, 1280]> input_307_cast_fp16 = reshape(shape = var_1735, x = var_1733_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_30_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205101056)))]; |
| tensor<fp16, [1280]> encoder_layers_30_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208377920)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_183_cast_fp16 = linear(bias = encoder_layers_30_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_30_self_attn_out_proj_weight_to_fp16, x = input_307_cast_fp16)[name = tensor<string, []>("linear_183_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_309_cast_fp16 = add(x = input_305_cast_fp16, y = linear_183_cast_fp16)[name = tensor<string, []>("input_309_cast_fp16")]; |
| tensor<int32, [1]> input_311_axes_0 = const()[name = tensor<string, []>("input_311_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_30_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208380544)))]; |
| tensor<fp16, [1280]> encoder_layers_30_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208383168)))]; |
| tensor<fp16, [1, 1500, 1280]> input_311_cast_fp16 = layer_norm(axes = input_311_axes_0, beta = encoder_layers_30_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_30_final_layer_norm_weight_to_fp16, x = input_309_cast_fp16)[name = tensor<string, []>("input_311_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_30_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208385792)))]; |
| tensor<fp16, [5120]> encoder_layers_30_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221493056)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_184_cast_fp16 = linear(bias = encoder_layers_30_fc1_bias_to_fp16, weight = encoder_layers_30_fc1_weight_to_fp16, x = input_311_cast_fp16)[name = tensor<string, []>("linear_184_cast_fp16")]; |
| tensor<string, []> input_313_mode_0 = const()[name = tensor<string, []>("input_313_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_313_cast_fp16 = gelu(mode = input_313_mode_0, x = linear_184_cast_fp16)[name = tensor<string, []>("input_313_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_30_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221503360)))]; |
| tensor<fp16, [1280]> encoder_layers_30_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_30_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234610624)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_185_cast_fp16 = linear(bias = encoder_layers_30_fc2_bias_to_fp16, weight = encoder_layers_30_fc2_weight_to_fp16, x = input_313_cast_fp16)[name = tensor<string, []>("linear_185_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_315_cast_fp16 = add(x = input_309_cast_fp16, y = linear_185_cast_fp16)[name = tensor<string, []>("input_315_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_191_axes_0 = const()[name = tensor<string, []>("hidden_states_191_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_31_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234613248)))]; |
| tensor<fp16, [1280]> encoder_layers_31_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234615872)))]; |
| tensor<fp16, [1, 1500, 1280]> hidden_states_191_cast_fp16 = layer_norm(axes = hidden_states_191_axes_0, beta = encoder_layers_31_self_attn_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_31_self_attn_layer_norm_weight_to_fp16, x = input_315_cast_fp16)[name = tensor<string, []>("hidden_states_191_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234618496)))]; |
| tensor<fp16, [1280]> encoder_layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237895360)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_186_cast_fp16 = linear(bias = encoder_layers_31_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_31_self_attn_q_proj_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor<string, []>("linear_186_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237897984)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = encoder_layers_31_self_attn_k_proj_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor<string, []>("linear_187_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241174848)))]; |
| tensor<fp16, [1280]> encoder_layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244451712)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_188_cast_fp16 = linear(bias = encoder_layers_31_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_31_self_attn_v_proj_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor<string, []>("linear_188_cast_fp16")]; |
| tensor<int32, [4]> var_1776 = const()[name = tensor<string, []>("op_1776"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1777_cast_fp16 = reshape(shape = var_1776, x = linear_186_cast_fp16)[name = tensor<string, []>("op_1777_cast_fp16")]; |
| tensor<int32, [4]> var_1779 = const()[name = tensor<string, []>("op_1779"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1780_cast_fp16 = reshape(shape = var_1779, x = linear_187_cast_fp16)[name = tensor<string, []>("op_1780_cast_fp16")]; |
| tensor<int32, [4]> var_1782 = const()[name = tensor<string, []>("op_1782"), val = tensor<int32, [4]>([1, 1500, 20, 64])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1783_cast_fp16 = reshape(shape = var_1782, x = linear_188_cast_fp16)[name = tensor<string, []>("op_1783_cast_fp16")]; |
| tensor<int32, [4]> value_perm_0 = const()[name = tensor<string, []>("value_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 1500, 20, 64]> mul_31_cast_fp16 = mul(x = var_1777_cast_fp16, y = var_6_to_fp16)[name = tensor<string, []>("mul_31_cast_fp16")]; |
| tensor<bool, []> matmul_31_transpose_y_0 = const()[name = tensor<string, []>("matmul_31_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_31_transpose_x_0 = const()[name = tensor<string, []>("matmul_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_190_perm_0 = const()[name = tensor<string, []>("transpose_190_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_191_perm_0 = const()[name = tensor<string, []>("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_191 = transpose(perm = transpose_191_perm_0, x = var_1780_cast_fp16)[name = tensor<string, []>("transpose_193")]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_190 = transpose(perm = transpose_190_perm_0, x = mul_31_cast_fp16)[name = tensor<string, []>("transpose_194")]; |
| tensor<fp16, [1, 20, 1500, 1500]> matmul_31_cast_fp16 = matmul(transpose_x = matmul_31_transpose_x_0, transpose_y = matmul_31_transpose_y_0, x = transpose_190, y = transpose_191)[name = tensor<string, []>("matmul_31_cast_fp16")]; |
| tensor<int32, []> softmax_31_axis_0 = const()[name = tensor<string, []>("softmax_31_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 20, 1500, 1500]> softmax_31_cast_fp16 = softmax(axis = softmax_31_axis_0, x = matmul_31_cast_fp16)[name = tensor<string, []>("softmax_31_cast_fp16")]; |
| tensor<bool, []> attn_output_transpose_x_0 = const()[name = tensor<string, []>("attn_output_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_transpose_y_0 = const()[name = tensor<string, []>("attn_output_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> value_cast_fp16 = transpose(perm = value_perm_0, x = var_1783_cast_fp16)[name = tensor<string, []>("transpose_195")]; |
| tensor<fp16, [1, 20, 1500, 64]> attn_output_cast_fp16 = matmul(transpose_x = attn_output_transpose_x_0, transpose_y = attn_output_transpose_y_0, x = softmax_31_cast_fp16, y = value_cast_fp16)[name = tensor<string, []>("attn_output_cast_fp16")]; |
| tensor<int32, [4]> var_1786_perm_0 = const()[name = tensor<string, []>("op_1786_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1788 = const()[name = tensor<string, []>("op_1788"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1786_cast_fp16 = transpose(perm = var_1786_perm_0, x = attn_output_cast_fp16)[name = tensor<string, []>("transpose_192")]; |
| tensor<fp16, [1, 1500, 1280]> input_317_cast_fp16 = reshape(shape = var_1788, x = var_1786_cast_fp16)[name = tensor<string, []>("input_317_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_layers_31_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244454336)))]; |
| tensor<fp16, [1280]> encoder_layers_31_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247731200)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_189_cast_fp16 = linear(bias = encoder_layers_31_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_31_self_attn_out_proj_weight_to_fp16, x = input_317_cast_fp16)[name = tensor<string, []>("linear_189_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_319_cast_fp16 = add(x = input_315_cast_fp16, y = linear_189_cast_fp16)[name = tensor<string, []>("input_319_cast_fp16")]; |
| tensor<int32, [1]> input_321_axes_0 = const()[name = tensor<string, []>("input_321_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layers_31_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247733824)))]; |
| tensor<fp16, [1280]> encoder_layers_31_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247736448)))]; |
| tensor<fp16, [1, 1500, 1280]> input_321_cast_fp16 = layer_norm(axes = input_321_axes_0, beta = encoder_layers_31_final_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layers_31_final_layer_norm_weight_to_fp16, x = input_319_cast_fp16)[name = tensor<string, []>("input_321_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_layers_31_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247739072)))]; |
| tensor<fp16, [5120]> encoder_layers_31_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260846336)))]; |
| tensor<fp16, [1, 1500, 5120]> linear_190_cast_fp16 = linear(bias = encoder_layers_31_fc1_bias_to_fp16, weight = encoder_layers_31_fc1_weight_to_fp16, x = input_321_cast_fp16)[name = tensor<string, []>("linear_190_cast_fp16")]; |
| tensor<string, []> input_323_mode_0 = const()[name = tensor<string, []>("input_323_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> input_323_cast_fp16 = gelu(mode = input_323_mode_0, x = linear_190_cast_fp16)[name = tensor<string, []>("input_323_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_layers_31_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260856640)))]; |
| tensor<fp16, [1280]> encoder_layers_31_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_31_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273963904)))]; |
| tensor<fp16, [1, 1500, 1280]> linear_191_cast_fp16 = linear(bias = encoder_layers_31_fc2_bias_to_fp16, weight = encoder_layers_31_fc2_weight_to_fp16, x = input_323_cast_fp16)[name = tensor<string, []>("linear_191_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1280]> input_325_cast_fp16 = add(x = input_319_cast_fp16, y = linear_191_cast_fp16)[name = tensor<string, []>("input_325_cast_fp16")]; |
| tensor<int32, [1]> encoder_output_axes_0 = const()[name = tensor<string, []>("encoder_output_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273966528)))]; |
| tensor<fp16, [1280]> encoder_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273969152)))]; |
| tensor<fp16, [1, 1500, 1280]> encoder_output_cast_fp16 = layer_norm(axes = encoder_output_axes_0, beta = encoder_layer_norm_bias_to_fp16, epsilon = var_4_to_fp16, gamma = encoder_layer_norm_weight_to_fp16, x = input_325_cast_fp16)[name = tensor<string, []>("encoder_output_cast_fp16")]; |
| tensor<int32, [3]> var_1819 = const()[name = tensor<string, []>("op_1819"), val = tensor<int32, [3]>([1, 375, 5120])]; |
| tensor<fp16, [1, 375, 5120]> input_327_cast_fp16 = reshape(shape = var_1819, x = encoder_output_cast_fp16)[name = tensor<string, []>("input_327_cast_fp16")]; |
| tensor<fp16, [3072, 5120]> projector_linear_1_weight_to_fp16 = const()[name = tensor<string, []>("projector_linear_1_weight_to_fp16"), val = tensor<fp16, [3072, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273971776)))]; |
| tensor<fp16, [3072]> linear_192_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_192_bias_0_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305429120)))]; |
| tensor<fp16, [1, 375, 3072]> linear_192_cast_fp16 = linear(bias = linear_192_bias_0_to_fp16, weight = projector_linear_1_weight_to_fp16, x = input_327_cast_fp16)[name = tensor<string, []>("linear_192_cast_fp16")]; |
| tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 375, 3072]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_192_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| tensor<fp16, [3072, 3072]> projector_linear_2_weight_to_fp16 = const()[name = tensor<string, []>("projector_linear_2_weight_to_fp16"), val = tensor<fp16, [3072, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305435328)))]; |
| tensor<fp16, [1, 375, 3072]> linear_193_cast_fp16 = linear(bias = linear_192_bias_0_to_fp16, weight = projector_linear_2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("linear_193_cast_fp16")]; |
| tensor<string, []> linear_193_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_193_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [1, 375, 3072]> audio_embeddings = cast(dtype = linear_193_cast_fp16_to_fp32_dtype_0, x = linear_193_cast_fp16)[name = tensor<string, []>("cast_131")]; |
| } -> (audio_embeddings); |
| } |
