arXiv Machine Learning

Mira: Memory-Efficient MoE Inference Using Adaptive Caching and Predictive Expert Staging

arXiv Computation and Language
Sep 7

Cache-Aware Joint Router Adaptation for Memory-Efficient MoE Inference

The paper introduces a cache‑aware post‑training framework for Mixture‑of‑Experts (MoE) models that jointly adapts the MoE backbone and lightweight auxiliary cache routers while keeping the native Top‑K expert‑selection rule. Two modes are proposed: Temporal Router, which predicts same‑layer reuse and retains experts for future tokens, and Spatio‑Temporal Router, which adds a Spatio Router that refines the temporal cache using the causal predecessor’s hidden state. Experiments on Qwen3 and GPT‑OSS across GSM8K, MATH, and CommonsenseQA show that Temporal Router improves cache hit rates and reduces expert‑weight traffic, while Spatio‑Temporal Router achieves the best load‑adjusted efficiency, outperforming strong prefetching baselines.

By Zhenhe Wu, Yaping Jin, Qinghua Xing, Hang Zhou, Wei He, Xianjie Wu, Xianfu Cheng, Jian Yang, Hanting Chen
arXiv AI
Jun 30

KernelSight-LM: A Kernel-Level LLM Inference Simulator

arXiv:2606. 28565v1 Announce Type: cross Abstract: As large language models (LLMs) move into production serving, practitioners must rapidly evaluate inference performance across diverse hardware, models, and serving parameters to meet cost and latency targets.

By Xiteng Yao, Taeho Kim, Hengzhi Pei, Xinle Liu, Kyle Ulrich, Leonard Lausen, Ashish Khetan, Xiang Song, George Karypis, Martin Herbordt
arXiv AI
Oct 31

Why Speculative Decoding Works Better Than Predicted on Sparse MoE Models

The paper investigates why speculative decoding (SD) performs better than expected on sparse mixture-of-experts (MoE) models. By evaluating 10 large MoE models, the authors find that recent fine‑grained models achieve higher speedups at low batch sizes, contrary to prior predictions of a concave speedup‑batch relationship. They attribute this to lower verification costs driven by temporal overlap of draft tokens and GPU execution dynamics, and develop a maximum‑entropy IPF framework that accurately predicts unique expert counts from routing overlap, closely matching real‑model behavior.

By Ekagra Ranjan, Komal K. Teru, Bharat Venkitesh, Acyr Locatelli
arXiv Machine Learning
Sep 14

Dynamic Expert Quantization for Scalable Mixture-of-Experts Inference

Dynamic Expert Quantization (DynaExq) is a runtime-aware mixed-precision serving system designed for single‑GPU Mixture‑of‑Experts (MoE) inference under a hard high‑bandwidth memory (HBM) envelope. It treats the problem as an online, budget‑constrained precision allocation task, keeping the most frequently used experts at higher precision while relegating the rest to low‑precision fallbacks. By estimating expert hotness from router traces and asynchronously promoting or demoting experts, DynaExq maintains a fully materialized expert set during the forward pass, improving accuracy and throughput compared to static post‑training quantization and offloading/prefetch baselines. whyItMatters":"DynaExq enables efficient deployment of large MoE models on memory‑limited GPUs by dynamically allocating precision based on runtime expert usage, thereby reducing memory footprint and latency while boosting accuracy and throughput."

By Kexin Chu, Dawei Xiang, Zixu Shen, Yiwei Yang, Zecheng Liu, Wei Zhang