OSS TanbouOn this page
Overview
PEFT adapts large pretrained models without updating every model parameter. LoRA and other adapter or prompt-tuning methods train a small parameter set, reducing memory and storage costs while integrating with Transformers, Diffusers, Accelerate, and the broader Hugging Face stack.
Features and best fit
Based on official documentation; not hands-on tested · Content checked:
Key features
Attach LoRA and other adapters while limiting trainable parameters
Configurations such as LoraConfig combine with get_peft_model to inject adapters into selected base-model modules so only a small fraction of parameters need training.
Sources: [1]
Store and switch adapter checkpoints separately from the base model
Fine-tuned adapters can be saved independently of the complete base model, loaded with PeftModel, and switched through Transformers integrations.
Sources: [1]
Integrate with Transformers, Diffusers, Accelerate, and related tooling
PEFT supports language-model workflows as well as diffusion models and distributed training or inference through Hugging Face ecosystem integrations.
Sources: [1]
Best fit
Fits repeated model adaptation where full fine-tuning is too expensive
It is useful when teams want to experiment across tasks or datasets without creating and storing a complete copy of a large base model for every fine-tuned variant.
Sources: [1]
Before adoption
Version 0.21.2 requires Python 3.10+ and PyTorch 1.13+
Package metadata for v0.21.2 requires Python >=3.10.0 and torch>=1.13.0 and depends on Transformers, Accelerate, Safetensors, and Hugging Face Hub. Check compatibility with the selected base model as well.
Version 0.21.2 fixes encoder-decoder compatibility with Transformers 5.18+
The v0.21.2 patch release fixes an issue that prevented encoder-decoder models from working with Transformers 5.18.0 and later, so affected environments should update the PEFT pin together with Transformers.
Sources: [2]
Official sources
- [1]PEFT v0.21.2 README(2026-10-04)
- [2]PEFT v0.21.2 release(2026-10-04)
- [3]PEFT v0.21.2 package metadata(2026-10-04)
- [4]PEFT v0.21.2 version metadata(2026-10-04)
- [5]PEFT Apache 2.0 license(2026-10-04)
Supplemental curator note
PEFT itself is Apache-2.0, but the base model and training dataset have separate licenses and terms. Check the upstream model conditions when distributing adapters.
Try it in 3 steps
- 1
Create a virtual environment for PEFT 0.21.2
Use Python 3.10+ and pin PEFT to the stable 0.21.2 release.
python3 -m venv peft-demo && . peft-demo/bin/activate && python -m pip install --upgrade pip && python -m pip install 'peft==0.21.2' - 2
Verify the PEFT version
Check the loaded library version without downloading a model.
. peft-demo/bin/activate && python -c 'import peft; print(peft.__version__)' - 3
Create a LoRA configuration without a model
Inspect the LoRA adapter configuration API without a GPU or base-model download.
. peft-demo/bin/activate && python -c 'from peft import LoraConfig, TaskType; c=LoraConfig(r=8,lora_alpha=16,task_type=TaskType.CAUSAL_LM); print(c.peft_type, c.r, c.lora_alpha)'
Growth
Growth trends · Last 30 days
21,751 Stars
Trend data is still being collected.
Development activity
Last 90 days · weekly
- Commits (last 30 days)
- 68
- Open PRs
- 51
Development activity is still being collected.
Built with
Categories and tags
Categories
GitHub data
GitHub dataView detailed GitHub data
GitHub Topics
- adapter
- diffusion
- llm
- parameter-efficient-learning
- python
- pytorch
- transformers
- lora
- fine-tuning
- peft
- Stars
- 21,751
- Forks
- 2,556
- Watchers
- 113
- Open issues
- 60
- Contributors
- 358
- Owner type
- Organization
- Primary language
- Python
- License
- Apache-2.0
- Repository last updated
- Oct 2, 2026
Related information
Write a related articleShare a guide or use case for this OSS in Markdown. Articles are published after administrator approval.
Explore next
- Mammoth842 Stars
3 shared tag(s) · 1 shared category(s) · Same language
Benchmark, reproduce, and extend 70+ continual-learning methods on one PyTorch framework
Python - TensorFlow200,702 Stars
3 shared tag(s) · 1 shared category(s)
machine-learning infrastructure with important Python and dependency changes in 2.21
C++ - Streamlit45,896 Stars
3 shared tag(s) · Same language
Turn Python scripts into interactive data apps with widgets, dataframes, charts, multipage navigation, and chat
Python - Langflow155,504 Stars
2 shared tag(s) · 1 shared category(s) · Same language
make AI workflows visible, then publish them as APIs and MCP tools
Python - Open WebUI153,981 Stars
2 shared tag(s) · 1 shared category(s) · Same language
combine local models, cloud providers, RAG, and tools in a self-hosted AI workspace
Python - LangChain147,451 Stars
2 shared tag(s) · 1 shared category(s) · Same language
compose models, tools, retrieval, and middleware into agents
Python
Report incorrect information
Tell us if any listing information is incorrect or outdated.