huggingface / trl

Train transformer language models with reinforcement learning.
http://hf.co/docs/trl
Apache License 2.0
10.13k stars 1.28k forks source link

🗃️ Use specified `data_collator` in `RLOOTrainer` and `PPOTrainer` #2360

Closed bartoszzuk closed 4 days ago

bartoszzuk commented 1 week ago

What does this PR do?

Fixes #2354

Before submitting

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR.

kashif commented 4 days ago

thanks @bartoszzuk perhaps it's better to set the self.data_collator to the default one if it is none and then use self.data_collator in the data loaders?

kashif commented 4 days ago

you might need to run make precommit in the root of the TRL to fix styling

HuggingFaceDocBuilderDev commented 4 days ago

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.