Can't train CG-BERT model about quasi-attention-absa HOT 2 CLOSED

Hi, Thanks for your inputs.

Yes, this is a TODO before. I was manually changing some lines of codes. This is due to the fact that CGBERT returns 2 data structs whereas QACGBERT returns more than 2 (the others were for analyses purposes), and we forgot to change CGBERT.

The fix just checked in. It is a one line change, you can refer to it and pull the master. Or you can simply change the line by yourself.

I attached logs from my manual testing:

06/04/2021 15:51:23 - INFO - util.train_helper - model = CGBERT
init_weight = True
06/04/2021 15:51:25 - INFO - util.train_helper - retraining with saved model.
100%|█████████████████████████████████████████████████████████████████████████████████████████████| 15220/15220 [00:03<00:00, 4500.98it/s]
06/04/2021 15:51:30 - INFO - util.train_helper - ***** Running training *****
06/04/2021 15:51:30 - INFO - util.train_helper - Num examples = 15220
06/04/2021 15:51:30 - INFO - util.train_helper - Batch size = 32
06/04/2021 15:51:30 - INFO - util.train_helper - Num steps = 14268
100%|███████████████████████████████████████████████████████████████████████████████████████████████| 4000/4000 [00:00<00:00, 4346.86it/s]
Epoch: 0%| | 0/30 [00:00<?, ?it/s/dfs/user/wuzhengx/workspace/Quasi-Attention-ABSA/code/util/optimization.py:146: UserWarning: This overload of add_ is deprecated:?, ?it/s]
add_(Number alpha, Tensor other)
Consider using one of the following signatures instead:
add_(Tensor other, *, Number alpha) (Triggered internally at /pytorch/torch/csrc/utils/python_arg_parser.cpp:1005.)
next_m.mul_(beta1).add_(1 - beta1, grad)

Iteration: 8%|█████▋ | 38/476

Happy debugging! Thanks!

from quasi-attention-absa.

Hi, @frankaging thanks for your feedback. I have just run the command for training CGBERT and all went well. Thanks

from quasi-attention-absa.