huggingface / trl

Train transformer language models with reinforcement learning.
http://hf.co/docs/trl
Apache License 2.0
10.18k stars 1.29k forks source link

Contributing new distillation related trainers #2361

Open YihanCao123 opened 1 week ago

YihanCao123 commented 1 week ago

Method description

Hi team, I have implemented some distillation based trainers, and would like contribute them to trl. Do you accept contributions on this or probably this is something already in progress? I see GKD has already been added into the trainer list, but some of the more basic distillation methods, as well as other knowledge distillation techniques, haven’t been included yet. I’d be happy to help expand if there’s interest. Thanks!

Open source status

Provide useful links for the implementation

No response

kashif commented 5 days ago

yes would welcome distillation trainers!