AI RESEARCH
When pre-training hurts LoRA fine-tuning: a dynamical analysis via single-index models
arXiv CS.LG
•
Pre-
Pre-