PEFT
Parameter-efficient fine-tuning methods behind one interface.
Resumen
PEFT collects the parameter-efficient fine-tuning methods - LoRA and its variants, prompt tuning, adapters - behind one interface so they can be attached to a pretrained model without rewriting training code. It solves a practical problem rather than a conceptual one: the methods exist in papers, and each arrives with its own implementation until somebody unifies them. It exists because most people adapting a model want the method, not the plumbing, and it is what made adapters interchangeable.
Grafo de tecnologías
Cómo se conecta PEFT con el resto del stack.
Aplicado a
- Fine-tuning technique Techniques PEFT used for Fine-tuning High ✓ source
- LoRA technique Techniques PEFT used for LoRA High ✓ source
- QLoRA technique Techniques PEFT used for QLoRA High ✓ source