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
- FIX Preserve CPT input token roles
- FIX Support MoE expert target_parameters with TP/EP
- Add a TiedLinear wrapper/proxy for lm_head when using tied embeddings
- FIX Healthcheck now also works for prompt learning
- `ensure_weight_tying=True` silently stops tying the LoRA weights when the model is in bf16/fp16
- LoRA on MoE experts (target_parameters) fails under tensor/expert parallelism
- Fix PeftMixedModel.disable_adapter state restoration
- FIX Apply OFT module_dropout to the rotation blocks again
- Default CPTConfig masks all input labels and returns NaN training loss
- Adapters are saved under transformers v5's renamed module paths, while `save_pretrained` writes base models in the original checkpoint layout
- Docs
- not yet supported