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
- LoRA-FA skips active adapter updates when a second adapter is present
- Cache the frozen LoRA-FA preconditioner to avoid repeated pseudoinverses
- Prompt tuning: batched generate output depends on left padding
- ENH Add generic PEFT loss and optmizer hook
- Nightly regression: bf16 state dict cases fail due to CPU kernel differences (not a PEFT bug)
- FIX DeFT, TinyLoRA and PSOFT second adapter overwriting fan_in_fan_out
- TST Speed up the Mixtral weight conversion tests on CPU
- TST Raise bf16 tolerance in state dict regression tests
- docs: improve LayerNorm Tuning documentation
- AdaMSS: forward crashes after casting the PEFT model's dtype (e.g. model.to(torch.bfloat16)) because the layer caches the base weight dtype at init
- Docs
- not yet supported