Your Experts Have Partners. There's HOPE for Keeping Them Together
Hugging Face Blog2712 字 (约 11 分钟)
85
HOPE方法通过二阶专家剪枝在保持MoE模型性能的同时减少50%内存占用,实验证明其在代理编码任务上优于现有方法6.1%。
入选理由:HOPE方法在50%剪枝率下实现6.1%性能提升,优于其他剪枝方法
FeaturedArticle#MoE#模型压缩#专家剪枝#Hugging Face#Amazon英文