Layer-Informed Fine-Tuning via Three-Stage Functional Segmentation of LLMs

Junning Shao, Siwei Wang, Zhixuan Fang

In recent years, the performance of large language models (LLMs) on reasoning tasks has been remarkable, even surpassing human capabilities on various benchmarks. However, there re…