The paper formalizes expert caching in MoE LLMs as layered paging, proves near-tight competitive lower bounds, and shows a layer-aware LRU variant beats standard LRU on real traces.
In: Proceedings of the 23rd ACM Symposium on Parallelism in Algorithms and Architectures (SPAA)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Cache Management for Mixture-of-Experts LLMs -- extended version
The paper formalizes expert caching in MoE LLMs as layered paging, proves near-tight competitive lower bounds, and shows a layer-aware LRU variant beats standard LRU on real traces.