1 paper · 1 filter
Xinyuan Song, Keyu Wang, PengXiang Li +2
Recent studies suggest that the deeper layers of Large Language Models (LLMs) contribute little to representation learning and can often be removed without significant performance…