peft
https://github.com/huggingface/peft
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
not yet supported1 Subscribers
Add a CodeTriage badge to peft
Help out
- Issues
- load_adapter leaves a selectable adapter after a safetensors read failure
- Warn when LoRA SFT trains on special tokens whose input embedding is all zero (silent failure on Qwen2.5-Coder tool calls)
- RandLoRA: wrong matrix multiplication in 8-bit/4-bit get_delta_weight() and broken multi-adapter rank handling
- C3ALinear.forward chains multiple active adapters through x instead of summing them
- LoHa `rank_dropout` can produce NaN when the keep-mask is all zeros
- IA3 `feedforward_modules=['fc']` also marks modules like `not_fc` as feedforward
- BOFT on MPS: safe_merge raises LinAlgError instead of ValueError when the adapter has a NaN
- Pass `token_type_ids` and `position_ids` in `PeftModelForQuestionAnswering`
- Add Context Tuning as a new PEFT method, recently presented at ICML 2026
- VeRA and PVeRA fail in forward() and merge() when adding a second adapter with a lower rank
- Docs
- not yet supported