search

huggingface / alignment-handbook

Robust recipes to align language models with human and AI preferences
https://huggingface.co/HuggingFaceH4
Apache License 2.0
4.54k stars 393 forks source link

Make max samples work again #80

Closed kirill-fedyanin closed 9 months ago

kirill-fedyanin commented 9 months ago

There are parameters max_train_samples and max_eval_samples in DataArguments, but if you put them in config, nothing changes. I look at the code, they don't appear anywhere beside logging.

This PR makes them behave as described in config help "For debugging purposes or quicker training, truncate the number of evaluation examples to this value if set.""

kirill-fedyanin commented 9 months ago

Ok, make sense, it much cleaner. I open another PR