1 paper · 1 filter
Mengxia Yu, De Wang, Qi Shan +2
Recent works have shown a surprising result: a small fraction of Large Language Model (LLM) parameter outliers are disproportionately important to the quality of the model. LLMs co…