For trillion-parameter MoE serving, the kernel page cache delivers oracle-competitive expert caching, so serving systems should let the kernel own eviction and keep model knowledge for admission and advice.
Del Mundo, Mohammad Rastegari, and Mehrdad Farajtabar
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.OS 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Who Should Own the Expert Cache? Kernel-Managed Tiering for Trillion-Parameter MoE Inference
For trillion-parameter MoE serving, the kernel page cache delivers oracle-competitive expert caching, so serving systems should let the kernel own eviction and keep model knowledge for admission and advice.