Hellor,
Thank you for sharing your code.
Have you tried to fine-tune the bert-like model during the fine-tuning?
If I understood well you don't use the tags when using the bert embeddings, did you try using both?
Also there is 2 minor changes for the char_lstm.py file to work :
Line 42 replace n_embed by n_word_embed and return None with embed to avoid changing the training code.
Thank you in advance.
Hellor,
Thank you for sharing your code.
Have you tried to fine-tune the bert-like model during the fine-tuning?
If I understood well you don't use the tags when using the bert embeddings, did you try using both?
Also there is 2 minor changes for the char_lstm.py file to work :
Line 42 replace n_embed by n_word_embed and return None with embed to avoid changing the training code.
Thank you in advance.