Hugging Face PEFT / TRL
Information
Hugging Face PEFT and TRL are core libraries commonly used as the default foundation for adapter-based fine-tuning
and alignment workflows.
PEFT provides efficient adaptation methods such as LoRA, QLoRA, and DoRA, while TRL adds trainers for
workflows such as SFT, DPO, GRPO, and PPO.
Many higher-level tools use these libraries underneath, which makes them important to understand even when you mainly work through other frameworks.
Common use cases
- adapter-based fine-tuning,
- supervised fine-tuning and preference optimization,
- custom Python training pipelines,
- and building your own LLM training stack on top of widely adopted libraries.
Practical note
If you want the most flexible and widely recognized base layer for modern open-model fine-tuning, PEFT and TRL are
often the most important libraries to learn first.