PapersBuilders10:26Chinese contentWUSH-KV: KV缓存量化新方法WUSH-KV用数据自适应变换降低KV缓存量化误差,2-bit性能超越OSCAR,适合长上下文推理场景。#WUSH-KV#KV cache#quantization#SGLangaarXiv cs.LG@Jiale Chen et al. (5)SourceRead laterReadWorth followingUsefulFollow WUSH-KV