huggingface/peft

Health v2 · maintenance only

Adoption not part of the score

21795 stars · 2624 forks observed · 2026-10-17

What it is AI-extracted, prompt v1, taxonomy v1, 2026-10-19, confidence 0.98

🤗 PEFT is a Python library that provides state‑of‑the‑art parameter‑efficient fine‑tuning methods for large pretrained models. It integrates with Transformers, Diffusers and Accelerate to simplify training and inference of adapters, LoRA, IA³ and other PEFT techniques.

Use cases

• Fine‑tune a large language model on a downstream task while using only a fraction of the model's parameters.
• Apply LoRA or adapter modules to a pretrained model to reduce GPU memory consumption.
• Train multiple task‑specific adapters for the same base model and switch between them at inference time.
• Add soft prompts or IA³ modules to a model for prompt‑engineering without full model retraining.
• Run distributed fine‑tuning of billion‑parameter models using DeepSpeed or FSDP with PEFT.
• Experiment with different PEFT methods quickly via the unified API.

When to choose

• You have limited GPU/VRAM resources but need to adapt large LLMs.
• You want to train many task‑specific models without storing full copies of the base model.
• You need a library that works out‑of‑the‑box with Hugging Face Transformers and Accelerate.
• You prefer a Python‑centric ecosystem with strong community support and frequent updates.

When to avoid

• Your workflow requires full‑model fine‑tuning for maximal performance on a very specialized task.
• You are using frameworks other than PyTorch (e.g., TensorFlow or JAX) without community adapters.
• You need a production‑grade inference server that does not rely on PEFT‑specific wrappers.

Facets

library · maturity active

llm-training llm-inference prompt-engineering artificial-intelligence large-language-models machine-learning python windows parameter-efficient-fine-tuning peft lora adapters transformers-integration accelerate-support adapter retrieval-augmented-generation linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="huggingface/peft")