huggingface / trl

Train transformer language models with reinforcement learning.
http://hf.co/docs/trl
Apache License 2.0
9.56k stars 1.19k forks source link

CI / `KTOTrainer`: Remove old tests #1750

Closed younesbelkada closed 3 months ago

younesbelkada commented 3 months ago

https://github.com/huggingface/trl/pull/1734 changed the logic for handling non desirable outputs from the KTOTrainer. Therefore we need to remove the corresponding tests to make the CI happy

cc @vwxyzjn @kashif

HuggingFaceDocBuilderDev commented 3 months 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.

vwxyzjn commented 3 months ago

The DPO trainer tests seem to fail for some reason. Maybe @kashif can take a look?

image