MoEZip: Routing-Aware KV Cache Compression for Sparse Mixture-of-Experts LLMs

Overview

arXiv:2610.30206v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) language models scale model capacity efficiently by activating a small subset of parameters per token. However, in long-context inference, memory remains a major bottleneck because the KV cache grows linearly with sequence length. Existing KV cache compression methods reduce this cost for dense transformers, but overlook a problem specific to MoE: KV eviction can change which experts are selected during answer generation, d

SpinGraph analysis pending — check back after processing.