Capability · Framework — fine-tuning

PEFT (Hugging Face)

PEFT (Parameter-Efficient Fine-Tuning) is the Hugging Face library that turned LoRA, QLoRA, and friends into a one-line import. Instead of updating every parameter of a 70B-parameter model, PEFT adds small trainable adapters (typically <1% of parameters) and freezes the rest, enabling fine-tuning on consumer GPUs. It's the foundation for downstream libraries like TRL, Axolotl, and Unsloth.

Framework facts

Category
fine-tuning
Language
Python
License
Apache-2.0
Repository
https://github.com/huggingface/peft

Install

pip install peft transformers accelerate bitsandbytes

Quickstart

from peft import LoraConfig, get_peft_model
from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained('meta-llama/Llama-3.2-3B', load_in_4bit=True)
config = LoraConfig(r=16, lora_alpha=32, target_modules=['q_proj','v_proj'], task_type='CAUSAL_LM')
model = get_peft_model(model, config)
model.print_trainable_parameters()

Alternatives

  • Axolotl — opinionated config-driven
  • Unsloth — 2-5x faster LoRA
  • TorchTune — PyTorch-native

Frequently asked questions

Is PEFT just for LoRA?

LoRA/QLoRA are the most popular, but PEFT also ships AdaLoRA, IA3, Prefix Tuning, Prompt Tuning, P-Tuning, OFT, DoRA, and VeRA — most research methods land here first.

PEFT or Unsloth?

Unsloth uses PEFT underneath but bolts on fused CUDA kernels for 2-5x speedup and half the memory. If you're GPU-poor, use Unsloth. For maximum flexibility or obscure adapter types, use PEFT directly.

Sources

  1. PEFT docs — accessed 2026-04-20
  2. PEFT GitHub — accessed 2026-04-20

Related

  • Axolotl — Axolotl is a config-driven fine-tuning framework for open-weight LLMs — write one YAML file describing dataset, model, and training hyperparameters, and Axolotl handles SFT, DPO, ORPO, LoRA, and full-parameter runs.
  • TRL (Transformer Reinforcement Learning) — TRL is Hugging Face's official library for post-training LLMs — supervised fine-tuning, PPO, DPO, ORPO, KTO, GRPO, and reward-model training, all built on Transformers and Accelerate.
  • Unsloth — Unsloth is a Python library that fine-tunes open-source LLMs (Llama, Mistral, Qwen, Gemma, Phi) 2-5x faster than HuggingFace defaults with 60-80% less memory, using custom Triton kernels and manual backprop.
  • All Frameworks — Browse the complete frameworks library
  • VSET — Programmes offered — B.Tech programs (CSE, AI&ML, AI&DS, Electronics, IIoT)