CoQuant: Joint Weight-Activation Subspace Projection for Mixed-Precision LLMs
CoQuant:混合精度LLM的联合权重-激活子空间投影
机构 * School of Internet of Things, Nanjing University of Posts and Telecommunications(南京邮电大学物联网学院) ; Amazon AGI(亚马逊人工智能)
AI总结 CoQuant通过联合建模权重和激活的子空间,提升混合精度LLM的量化效果,在WikiText perplexity和零样本常识推理准确率上优于现有方法。
Comments 14 pages, 3 figures