Error when testing BertAbs model

[tetsu9923]

Hi, thank you for your great work!

I encountered the following error in predictor.py` when trying to test BertAbs model.

python train.py -task abs -mode test -test_from ../models/abs/model_step_8000.pt -batch_size 3000 -test_batch_size 500 -bert_data_path ../bert_data/bert.pt_data -log_file ../logs/test_abs -model_path ../models/abs -sep_optim true -use_interval true -visible_gpus 0 -max_pos 512 -max_length 200 -alpha 0.95 -min_length 50 -result_path ../logs/abs

Traceback (most recent call last): File "train.py", line 135, in test_abs(args, device_id, cp, step) File "/home/t-kasanishi/PreSumm/src/train_abstractive.py", line 225, in test_abs predictor.translate(test_iter, step) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 149, in translate batch_data = self.translate_batch(batch) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 218, in translate_batch min_length=self.min_length) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 333, in _fast_translate_batch [alive_seq.index_select(0, select_indices), RuntimeError: "index_select_out_cuda_impl" not implemented for 'Float'

Below is a shape and contents of select_indices and alive_seq. FloatTensor cannot be used for the argument of .index_select() (only IntTensor or LongTensor is suitable). How should I fix it? Thank you!

select_indices.shape: torch.Size([30])
select_indices: tensor([ 0.0654, 0.0655, 0.0657, 0.0340, 0.0674, 5.0654, 5.0655, 5.0657,
5.0340, 5.0674, 10.0654, 10.0657, 10.0655, 10.0340, 10.0657, 15.0654,
15.0655, 15.0657, 15.0340, 15.0674, 20.0654, 20.0655, 20.0657, 20.0340,
20.0657, 25.0654, 25.0655, 25.0657, 25.0340, 25.0674], device='cuda:0')
alive_seq.shape: torch.Size([30, 1])
alive_seq: tensor([[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1], [1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1],
[1], [1]], device='cuda:0')

[wbchief]

Hi, thank you for your great work!

I encountered the following error in predictor.py` when trying to test BertAbs model.

python train.py -task abs -mode test -test_from ../models/abs/model_step_8000.pt -batch_size 3000 -test_batch_size 500 -bert_data_path ../bert_data/bert.pt_data -log_file ../logs/test_abs -model_path ../models/abs -sep_optim true -use_interval true -visible_gpus 0 -max_pos 512 -max_length 200 -alpha 0.95 -min_length 50 -result_path ../logs/abs

Traceback (most recent call last): File "train.py", line 135, in test_abs(args, device_id, cp, step) File "/home/t-kasanishi/PreSumm/src/train_abstractive.py", line 225, in test_abs predictor.translate(test_iter, step) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 149, in translate batch_data = self.translate_batch(batch) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 218, in translate_batch min_length=self.min_length) File "/home/t-kasanishi/PreSumm/src/models/predictor.py", line 333, in _fast_translate_batch [alive_seq.index_select(0, select_indices), RuntimeError: "index_select_out_cuda_impl" not implemented for 'Float'

Below is a shape and contents of select_indices and alive_seq. FloatTensor cannot be used for the argument of .index_select() (only IntTensor or LongTensor is suitable). How should I fix it? Thank you!

select_indices.shape: torch.Size([30]) select_indices: tensor([ 0.0654, 0.0655, 0.0657, 0.0340, 0.0674, 5.0654, 5.0655, 5.0657, 5.0340, 5.0674, 10.0654, 10.0657, 10.0655, 10.0340, 10.0657, 15.0654, 15.0655, 15.0657, 15.0340, 15.0674, 20.0654, 20.0655, 20.0657, 20.0340, 20.0657, 25.0654, 25.0655, 25.0657, 25.0340, 25.0674], device='cuda:0') alive_seq.shape: torch.Size([30, 1]) alive_seq: tensor([[1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1], [1]], device='cuda:0')

topk_beam_index = topk_ids.div(vocab_size)

        topk_beam_index = topk_ids // vocab_size

[Dengguty]

请问解决了吗？我也遇到了这个问题