Reproducible Evaluation of MoE Expert Caching: Replay Semantics, Workload Contamination, and Operating Regimes

[2608.07911] Reproducible Evaluation of MoE Expert Caching: Replay Semantics, Workload Contamination, and Operating Regimes Skip to main content Search Submit Donate Log in Computer Science > Machine Learning arXiv:2608.07911 (cs) [Submitted on 8 Aug 2026 ( v1 ), last revised 25 Aug 2026 (this version, v4)] Title: Reproducible Evaluation of MoE Expert Caching: Replay Semantics, Workload Contamination, and Operating Regimes Authors: Yu Zhang View PDF HTML (experimental) Abstract: Mixture-of-Experts (MoE) models have outgrown accelerator memory, and offloading expert weights to host memory is now standard. This makes expert cache management an attractive lever: a policy that raised the hit rate would cut expert traffic per token. Evaluating that is a measurement problem, and we find the measurement fragile. With a trace-driven, event-atomic simulator over three MoE models (40, 64, 128 experts), we isolate three evaluation axes that change conclusions, not just numbers. Replay semantics: under a fused-event traffic contract, an inconsistent per-access replay inflates recency-based policies by 27-29% while leaving frequency-based and static ones within 4%, inverting the policy ranking. Workload contamination: probe sets using one instruction template per category produce verbatim-identical generation prefixes; a matched-pair rendering intervention moves the measured early-window effect by 19.4-31.9 points and reverses which workloads look most cache-friendly. Operating regimes: normalized miss fractions do not transfer across models, so the per-step expert union relative to per-layer capacity must be reported -- yet permuting only the temporal order of an identical event stream moves the offline-optimal gap from 44.9% to 30.8%, so it is not sufficient. Corrected, a stable gap to the offline optimum remains (44.2-45.9% over 13 frozen workload compositions). A forced-admission oracle attributes 84.3-96.6% of it to knowing which resident expert is used furthest in the future. A causal next-use predictor, used as an eviction rule, recovers -11.4% of the gap; it picks an optimal victim 3.4% of the time, against 2.4% for a random resident block and 20.6-22.1% for LRU and LFRU. Our position is narrow: in our evaluated settings a large offline-optimal gap substantially overstates the gains recovered by representative lightweight causal mechanisms. Comments: 39 pages, 4 figures, 21 tables. Measurement and scoped negative-results study. Artifacts and SHA-256 manifests: this https URL and this https URL . v3: title changed; the generative-AI declaration now names both systems used. v4: metadata only; manuscript unchanged from v3 Subjects: Machine Learning (cs.LG) ; Performance (cs.PF) ACM classes: C.4; D.4.2; I.2.6 Cite as: arXiv:2608.07911 [cs.LG] (or arXiv:2608.07911v4 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2608.07911 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Yu Zhang [ view email ] [v1] Sat, 8 Aug 2026 04:29:20 UTC (77 KB) [v2] Thu, 13 Aug 2026 06:19:20 UTC (78 KB) [v3] Sat, 22 Aug 2026 09:15:12 UTC (78 KB) [v4] Tue, 25 Aug 2026 03:22:31 UTC (78 KB) Full-text links: Access Paper: View PDF HTML (experimental) TeX Source view license Current browse context: cs.LG < prev | next > new | recent | 2026-08 Change to browse by: cs cs.PF References & Citations NASA ADS Google Scholar Semantic Scholar export BibTeX citation Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer ( What is the Explorer? ) Connected Papers Toggle Connected Papers ( What is Connected Papers? ) Litmaps Toggle Litmaps ( What is Litmaps? ) scite.ai Toggle scite Smart Citations ( What are Smart Citations? ) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv ( What is alphaXiv? ) Links to Code Toggle CatalyzeX Code Finder for Papers ( What is CatalyzeX? ) DagsHub Toggle DagsHub ( What is DagsHub? ) GotitPub Toggle Gotit.pub ( What is GotitPub? ) Huggingface Toggle Hugging Face ( What is Huggingface? ) ScienceCast Toggle ScienceCast ( What is ScienceCast? ) Demos Demos Replicate Toggle Replicate ( What is Replicate? ) Spaces Toggle Hugging Face Spaces ( What is Spaces? ) Spaces Toggle TXYZ.AI ( What is TXYZ.AI? ) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower ( What are Influence Flowers? ) Core recommender toggle CORE Recommender ( What is CORE? ) IArxiv recommender toggle IArxiv Recommender ( What is IArxiv? ) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs . Which authors of this paper are endorsers? | Disable MathJax ( What is MathJax? ) We gratefully acknowledge support from our major funders , member institutions , and all contributors. About · Help · Contact · Subscribe · Copyright · Privacy · Accessibility · Operational Status (opens in new tab) Major funding support from

Related documents

• Reproducible Evaluation of MoE Expert Caching: Replay Semantics, Workload Contamination, and Operating Regimes#670737
• Trillion-Parameter MoE in a Box: Decoupling Memory Provisioning with High-Bandwidth Flash#1109618
• Analytical Resource Management for Fine-grained MoE Computation-Communication Overlap#984451
• BASE: Batch-Aware Selection of Experts Using Predicted Removal Error for Efficient MoE Decoding#1131064
• Algorithms for rescheduling jobs with a LIFO buffer to minimize the weighted number of late jobs#274657