search

allenai / reward-bench

RewardBench: the first evaluation tool for reward models.
https://huggingface.co/spaces/allenai/reward-bench
Apache License 2.0
375 stars 47 forks source link

Clean up / enhance DPO code #82

Closed natolambert closed 3 months ago

natolambert commented 6 months ago
  1. Make it so you can run inference over individual text prompts (rather than chosen + rejected)
  2. Clean up nograd/detach (see https://twitter.com/shxf0072/status/1771220126655811610), but should be pretty obvious
  3. Add per-model log prob computation.
natolambert commented 3 months ago

Closing because stale. Feel free to reopen.