Replace old generation_config.json with latest.

#8

The old generation_config.json does not specify <|eot_id|> as the stop token. This makes the model incapable of stopping when running it using transformers, which leads to low scores on the leaderboard V2.0 evaluation. (More info here: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/815 )

Thanks a lot @Dampfinchen !

mlabonne changed pull request status to merged

Sign up or log in to comment