Debian Question posted in
The oficial documentation can be found here.

Loading saved NER transformers model causes AttributeError? – Debian

I have trained and saved some NER models using

torch.save(model)

I need to load these model files (extension .pt) for evaluation using

torch.load('PATH_TO_MODEL.pt')

And I get the following error: 'BertConfig' object has no attribute 'return_dict'

For the same, I updated my transformer package to the latest one, but the error persists.

This is the stack trace:

Traceback (most recent call last):
File "/home/systematicReviews/train_mtl_3.py", line 523, in <module>
test_loss, test_cr, test_cr_fine = evaluate_i(test_model, optimizer, scheduler, validation_dataloader, args, device)
File "/home/systematicReviews/train_mtl_3.py", line 180, in evaluate_i
e_loss_coarse, e_output, e_labels, e_loss_fine, e_f_output, e_f_labels, mask, e_cumulative_loss  = defModel(args, e_input_ids, attention_mask=e_input_mask, P_labels=e_labels, P_f_labels=e_f_labels)
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
result = self.forward(*input, **kwargs)
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 150, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
result = self.forward(*input, **kwargs)
File "/home/systematicReviews/models/mtl/model.py", line 122, in forward
attention_mask = attention_mask
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
result = self.forward(*input, **kwargs)
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/transformers/modeling_bert.py", line 784, in forward
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
File "/home/anaconda3/envs/systreviewclassifi/lib/python3.6/site-packages/transformers/configuration_utils.py", line 219, in use_return_dict
return self.return_dict and not self.torchscript
AttributeError: 'BertConfig' object has no attribute 'return_dict'

Here is some more information about my system:

- `transformers` version: 3.1.0
- Platform: Linux-4.4.0-186-generic-x86_64-with-debian-stretch-sid
- Python version: 3.6.9
- PyTorch version (GPU?): 1.3.1 (True)
- Tensorflow version (GPU?): not installed (NA)
- Using GPU in script?: Yes
- Using distributed or parallel set-up in script?: No

It worked pretty fine until now, but suddenly this bug appears. Any help or hint is appreciated.

Tags:

2

Answers


  1. Chosen as BEST ANSWER
    0

    To save model dictionary rather than an entire model is slightly different. Instead of torch.save(model) use torch.save('path_to_the_model/model.pth') and load using torch.load('path_to_the_model/model.pth').


  2. 0

    Try to save your model with model.save_pretrained(output_dir). Then you can load your model with model = *.from_pretrained(output_dir) where * is the model class (e.g. BertForTokenClassification).

    Login or Signup to reply.
Please signup or login to give your own answer.

Back To Top
Search