SMoE substitutes low-importance experts with cached similar ones in MoE inference on edge devices to achieve 48% lower decoding latency and over 60% cache hit rate with nearly lossless accuracy.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SMoE: An Algorithm-System Co-Design for Pushing MoE to the Edge via Expert Substitution
SMoE substitutes low-importance experts with cached similar ones in MoE inference on edge devices to achieve 48% lower decoding latency and over 60% cache hit rate with nearly lossless accuracy.