training

LoRA

Low-Rank Adaptation—an efficient fine-tuning method that updates small adapter weights instead of the full model.