Klotski pipelines multiple batches of MoE inference, prefetches only likely-used 'hot' experts, and reorders expert computations to overlap I/O with compute, claiming up to 85.12x throughput gains.
Title resolution pending
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
-
Klotski: Efficient Mixture-of-Expert Inference via Expert-Aware Multi-Batch Pipeline
Klotski pipelines multiple batches of MoE inference, prefetches only likely-used 'hot' experts, and reorders expert computations to overlap I/O with compute, claiming up to 85.12x throughput gains.