aiqbench analysis
This research addresses the critical memory bottleneck in Mixture-of-Experts (MoE) inference by introducing a cache-aware post-training framework that jointly optimizes the model backbone and auxiliary routers. By predicting expert reuse patterns, the proposed Temporal and Spatio-Temporal routers significantly reduce GPU weight transfer traffic without altering the native Top-K selection logic at inference time. The results demonstrate substantial efficiency gains on large models like Qwen3, offering a practical path to scaling MoE architectures within constrained memory budgets.
Cache-Aware Joint Router Adaptation for Memory-Efficient MoE Inference
arXiv:2609.04895v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) models activate only a small subset of experts per token, but the full expert set often exceeds GPU memory, causing repeated weight transfers during decoding. We formulate expert-cache management as a model-side algorithmic problem and propose a cache-aware post-training framework that jointly adapts the MoE backbone and lightweight auxiliary cache routers while preserving the native Top-K expert-selection rule at inference. Its update-only mode, Temporal Router, predicts same-layer reuse and retains experts for future tokens without proactive loading. The full Spatio-Temporal Router adds a Spatio Router that uses the causal predecessor's hidden state to refine the temporal cache before target-layer access. We evaluate both modes on Qwen3 and GPT-OSS across GSM8K, MATH, and CommonsenseQA. Temporal Router consistently improves cache hit rate and reduces expert-weight traffic over matched LM-only baselines. On Qwen3, Spatio-Temporal Router achieves the best load-adjusted efficiency across three tasks, improving adjusted hit rate by 1.15--18.03 points and reducing traffic by 4.6--53.3% relative to the strongest evaluated prefetching baseline; results on GPT-OSS are competitive but task-dependent. An auxiliary-only ablation preserves baseline accuracy but yields modest cache gains, whereas joint post-training produces larger improvements. Sensitivity analyses show that cache capacity controls transfer demand, while the refinement budget governs the trade-off between pre-access coverage and proactive traffic.
Abstract:Mixture-of-Experts (MoE) models activate only a small subset of experts per token, but the full expert set often exceeds GPU memory, causing repeated weight transfers during decoding. We formulate expert-cache management as a model-side algorithmic problem and propose a cache-aware post-training framework that jointly adapts the MoE backbone and lightweight auxiliary cache routers while preserving the native Top-K expert-selection rule at inference. Its update-only mode, Temporal Router, predicts same-layer reuse and retains experts for future tokens without proactive loading. The full Spatio-Temporal Router adds a Spatio Router that uses the causal predecessor's hidden state to refine the temporal cache before target-layer access. We evaluate both modes on Qwen3 and GPT-OSS across GSM8K, MATH, and CommonsenseQA. Temporal Router consistently improves cache hit rate and reduces expert-weight traffic over matched LM-only baselines. On Qwen3, Spatio-Temporal Router achieves the best load-adjusted efficiency across three tasks, improving adjusted hit rate by 1.15--18.03 points and reducing traffic by 4.6--53.3% relative to the strongest evaluated prefetching baseline; results on GPT-OSS are competitive but task-dependent. An auxiliary-only ablation preserves baseline accuracy but yields modest cache gains, whereas joint post-training produces larger improvements. Sensitivity analyses show that cache capacity controls transfer demand, while the refinement budget governs the trade-off between pre-access coverage and proactive traffic.
Computation and Language (cs.CL)
arXiv:2609.04895 [cs.CL]
(or arXiv:2609.04895v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.04895 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 302 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
[AINews] Modal’s $1B ARR, $16B Series D makes it NYC’s newest AI Infra Decacorn, 3x usage and 3x valuation in 5 months, 15x in 1 yr
Latent Space · Oct 15, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
Do Dynamic Routers Need Memory? HeRo: History-Aware Routing for Efficient LLM Inference
arXiv · Sep 9, 2026