fhamborg/NewsMTSC

Model update

Closed this issue · 3 comments

Hi, I recent rework the LCF-BERT model, now it is much faster and stable. If you have interest to update the model, please see PyABSA.

Cheers,

Yang Heng

Hi, this is interesting :-) Which of the models listed at https://github.com/yangheng95/PyABSA#model-support are you referring to? Is it still called LCF-BERT and thus this file? https://github.com/yangheng95/PyABSA/blob/release/pyabsa/module/apc/models/lcf_bert.py

I recommend to use fast-lcf-bert/fast-lcfs-bert, which adopts an effcient lcf vec calculation method and use the global context features output to learn local cotnext features (means halfing the time every step).

Thanks again for the hint. We can surely add the updated version. Would you be willing to create a PR with the updated variant?