princeton-nlp / LM-BFF

[ACL 2021] LM-BFF: Better Few-shot Fine-tuning of Language Models https://arxiv.org/abs/2012.15723

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ValueError Some specified arguments are not used by the HfArgumentParser

demongolem-biz opened this issue · comments

When running the run.py in the Quick Start session under Run LM-BFF, I get

raise ValueError(f"Some specified arguments are not used by the HfArgumentParser: {remaining args}"

I believe it is complaining about --evaluate_during_training. Is this deprecated? Should this simply be removed, or is there a replacement? I have had to fix the paths to Bert and Roberta in the imports so I think this just has to do with the evolution of transformers since this was originally published.

Hi,

It might be due to different versions of transformers. Can you check your installed tranformers version? The current code is designed for 3.4.0 transformers.

In this case --evaluation_strategy epoch would seem to be the correct replacement. I would agree that this would work with transformers 3.4.0. In the quick start section, it reads "but it should work with other versions". I think this part should be deleted and the assertion should read this requires transformers 3.4. As it is, when I replace --evaluate_during_training, there are other errors waiting behind which also result from mismatched version.

Good point, I'll edit the README.