Fei Wang

Leaner Training, Lower Leakage: Revisiting Memorization in LLM Fine-Tuning with LoRA

It is demonstrated that LoRA significantly reduces memorization risks compared to full fine-tuning, while still maintaining strong task performance.