systems-research

Tag

Cards List
#systems-research

COPE: Composable Performance Emulation for Mixture-of-Experts Offloading Policies

arXiv cs.LG ↗ · 2026-10-28 Cached

COPE: Composable Performance Emulation for Mixture-of-Experts Offloading Policies——上海交大IPADS提出的一个框架,通过Policy Matrix和Virtual Heterogeneous Machine在运行时集成前预测并选出最优的MoE专家卸载策略,将最快策略选择准确率提升24%–100%,性能预测误差降低1.7×–6.8×,并减少最多74%的策略实现工作量。

0 favorites 0 likes
#systems-research

TokenRouter: Efficient Serving System for Token-Level LLM Routing

Hugging Face Daily Papers ↗ · 2026-10-08 Cached

TokenRouter 是一个面向 token 级 LLM 路由的高效推理服务系统,采用「请求中心编程、模型中心执行」架构与延迟批调度策略,实现较现有系统 2.01–64.15 倍的解码吞吐提升。

0 favorites 0 likes
#systems-research

TierKV: Long-Context On-Device LLMs via Predictive Multi-Tier KV Caching

arXiv cs.LG ↗ · 2026-09-21 Cached

TierKV proposes a predictive multi-tier KV caching framework to optimize memory usage and throughput for long-context LLMs on mobile devices, achieving significant performance improvements with minimal accuracy degradation.

0 favorites 0 likes
← Back to home

Submit Feedback