Bert-Multi-Label-Text-Classification
Bert-Multi-Label-Text-Classification copied to clipboard
Updating to pytorch-transformers
Fix to #14. I believe that this should update the things that have changed from the migration from pretrainedbert to pytorch-transformers. This update runs fine for me using the toxic Kaggle dataset however, I think there may be some other things in trainer.py that I have missed.
Also in bert_fine.py, I have commented out output_all_encoded_layers in the forward function as I'm not sure how to include this
Thank you! By the way, focal loss is better than BCE loss, especially for hundreds of labels