Trace-driven evaluation of mixture-of-experts expert caching is fragile in at least four ways; after correction, most of the offline-optimal gap in the tested settings depends on future knowledge that simple causal predictors fail to recover.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
When Does Trace-Driven Evaluation Mislead MoE Expert Caching? Replay Semantics, Workload Contamination, and Operating Regimes
Trace-driven evaluation of mixture-of-experts expert caching is fragile in at least four ways; after correction, most of the offline-optimal gap in the tested settings depends on future knowledge that simple causal predictors fail to recover.