AI RESEARCH

When pre-training hurts LoRA fine-tuning: a dynamical analysis via single-index models

arXiv CS.LG

Pre-