RuntimeError: 'lengths' argument should be a 1D CPU int64 tensor, but got 1D cuda:0 Long tensor

[robi56]

python train.py

hyper-patameters: HParams(vocab_size=7024, pad_idx=0, bos_idx=3, emb_size=256, hidden_size=512, context_size=512, latent_size=256, factor_emb_size=64, n_class1=3, n_class2=2, key_len=4, sens_num=4, sen_len=9, poem_len=30, batch_size=128, drop_ratio=0.15, weight_decay=0.00025, clip_grad_norm=2.0, max_lr=0.0008, min_lr=5e-08, warmup_steps=6000, ndis=3, min_tr=0.85, burn_down_tr=3, decay_tr=6, tau_annealing_steps=6000, min_tau=0.01, rec_warm_steps=1500, noise_decay_steps=8500, log_steps=200, sample_num=1, max_epoches=12, save_epoches=3, validate_epoches=1, fbatch_size=64, fmax_epoches=3, fsave_epoches=1, vocab_path='../corpus/vocab.pickle', ivocab_path='../corpus/ivocab.pickle', train_data='../corpus/semi_train.pickle', valid_data='../corpus/semi_valid.pickle', model_dir='../checkpoint/', data_dir='../data/', train_log_path='../log/mix_train_log.txt', valid_log_path='../log/mix_valid_log.txt', fig_log_path='../log/', corrupt_ratio=0.1, dae_epoches=10, dae_batch_size=128, dae_max_lr=0.0008, dae_min_lr=5e-08, dae_warmup_steps=4500, dae_min_tr=0.85, dae_burn_down_tr=2, dae_decay_tr=6, dae_log_steps=300, dae_validate_epoches=1, dae_save_epoches=2, dae_train_log_path='../log/dae_train_log.txt', dae_valid_log_path='../log/dae_valid_log.txt', cl_batch_size=64, cl_epoches=10, cl_max_lr=0.0008, cl_min_lr=5e-08, cl_warmup_steps=800, cl_log_steps=100, cl_validate_epoches=1, cl_save_epoches=2, cl_train_log_path='../log/cl_train_log.txt', cl_valid_log_path='../log/cl_valid_log.txt') please check the hyper-parameters, and then press any key to continue >ok ok dae pretraining... layers.embed.weight torch.Size([7024, 256]) layers.encoder.rnn.weight_ih_l0 torch.Size([1536, 256]) layers.encoder.rnn.weight_hh_l0 torch.Size([1536, 512]) layers.encoder.rnn.bias_ih_l0 torch.Size([1536]) layers.encoder.rnn.bias_hh_l0 torch.Size([1536]) layers.encoder.rnn.weight_ih_l0_reverse torch.Size([1536, 256]) layers.encoder.rnn.weight_hh_l0_reverse torch.Size([1536, 512]) layers.encoder.rnn.bias_ih_l0_reverse torch.Size([1536]) layers.encoder.rnn.bias_hh_l0_reverse torch.Size([1536]) layers.decoder.rnn.weight_ih_l0 torch.Size([1536, 512]) layers.decoder.rnn.weight_hh_l0 torch.Size([1536, 512]) layers.decoder.rnn.bias_ih_l0 torch.Size([1536]) layers.decoder.rnn.bias_hh_l0 torch.Size([1536]) layers.word_encoder.rnn.weight_ih_l0 torch.Size([256, 256]) layers.word_encoder.rnn.weight_hh_l0 torch.Size([256, 256]) layers.word_encoder.rnn.bias_ih_l0 torch.Size([256]) layers.word_encoder.rnn.bias_hh_l0 torch.Size([256]) layers.word_encoder.rnn.weight_ih_l0_reverse torch.Size([256, 256]) layers.word_encoder.rnn.weight_hh_l0_reverse torch.Size([256, 256]) layers.word_encoder.rnn.bias_ih_l0_reverse torch.Size([256]) layers.word_encoder.rnn.bias_hh_l0_reverse torch.Size([256]) layers.out_proj.weight torch.Size([7024, 512]) layers.out_proj.bias torch.Size([7024]) layers.map_x.mlp.linear_0.weight torch.Size([512, 768]) layers.map_x.mlp.linear_0.bias torch.Size([512]) layers.context.conv.weight torch.Size([512, 512, 3]) layers.context.conv.bias torch.Size([512]) layers.context.linear.weight torch.Size([512, 1024]) layers.context.linear.bias torch.Size([512]) layers.dec_init_pre.mlp.linear_0.weight torch.Size([506, 1536]) layers.dec_init_pre.mlp.linear_0.bias torch.Size([506]) params num: 31 building data for dae... 193461 34210 train batch num: 1512 valid batch num: 268 Traceback (most recent call last): File "train.py", line 79, in main() File "train.py", line 74, in main pretrain(mixpoet, tool, hps) File "train.py", line 30, in pretrain dae_trainer.train(mixpoet, tool) File "/content/MixPoet/codes/dae_trainer.py", line 137, in train self.run_train(mixpoet, tool, optimizer, logger) File "/content/MixPoet/codes/dae_trainer.py", line 83, in run_train batch_keys, batch_poems, batch_dec_inps, batch_lengths) File "/content/MixPoet/codes/dae_trainer.py", line 58, in run_step mixpoet.dae_graph(keys, poems, dec_inps, lengths) File "/content/MixPoet/codes/graphs.py", line 337, in dae_graph _, poem_state0 = self.computer_enc(poems, self.layers['encoder']) File "/content/MixPoet/codes/graphs.py", line 232, in computer_enc enc_outs, enc_state = encoder(emb_inps, lengths) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1051, in _call_impl return forward_call(*input, **kwargs) File "/content/MixPoet/codes/layers.py", line 59, in forward input_lens, batch_first=True, enforce_sorted=False) File "/usr/local/lib/python3.7/dist-packages/torch/nn/utils/rnn.py", line 249, in pack_padded_sequence _VF._pack_padded_sequence(input, lengths, batch_first) RuntimeError: 'lengths' argument should be a 1D CPU int64 tensor, but got 1D cuda:0 Long tensor

[hiroki-chen]

You can change the following

packed = torch.nn.utils.rnn.pack_padded_sequence(embed_inps,
                input_lens, batch_first=True, enforce_sorted=False)

in file layer.py line 56 to

packed = torch.nn.utils.rnn.pack_padded_sequence(embed_inps,
                input_lens.cpu(), batch_first=True, enforce_sorted=False)

I think this is a bug introduced when you are using a newer version of PyTorch.