1
LoCO: Low-rank Compositional Rotation Fine-tuning
新方法LoCO通过低秩组合正交微调，在保持预训练几何结构的同时实现高效参数调整，为NLP和计算机视觉大模型适配提供新思路。
arXiv:2605.15916v1 Announce Type: cross Abstract: Parameter-efficient fine-tuning (PEFT) has emerged as an critical technique for adapting large-scale…