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
- [docs] Clarify AdaLoRA training phases and update_and_allocate
- FIX adapter dropout state for explicit `inference_mode=True`
- FSDP2 with `reshard_after_forward=False`: `disable_adapter()` leaves LoRA frozen and `merge_adapter()` is lost at reshard
- `all-linear` skips dense and shared-expert MLP layers on MoE models converted for transformers v5
- `save_pretrained` fails on FSDP2-sharded adapters: the DTensor gather only runs when `_tp_size` is set
- LoRA on an FSDP2-sharded model gets the local shard shape when the mesh dimension is not named `fsdp`
- Adapter loading never warns about unexpected keys, and `strict=True` is silently ignored
- `modules_to_save` with a persistent buffer raises KeyError on a params-only `state_dict`
- FIX Match hotswap adapters by exact name
- DoRA convolution dropout path ignores non-zero padding_mode during training
- Docs
- not yet supported