🧪 Test?View on arXiv
MoEZip: Routing-Aware KV Cache Compression for Sparse Mixture-of-Experts LLMs
Author1, Author2, Author3, Author4, Author5
Mixture-of-ExpertsKV Cache CompressionLong-Context Inference
2610.30206
Builder Relevance
2h ago80%
Abstract
MoEZip introduces a routing-aware KV cache compression method for sparse Mixture-of-Experts language models to improve memory efficiency during long-context inference.
Reality Card
Core Claim
MoEZip achieves robust performance under aggressive cache compression, improving state-of-the-art benchmarks by significant margins.
Method / Result
+24 and +20 points on average on LongBench and RULER benchmarks at a KV retention ratio of 10%
Limitations
The method's effectiveness may vary based on the specific architecture of the MoE model used.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.