Reproduce “MAServeBench: Benchmarking Multi-Agent LLM Serving Systems with Hierarchical Collaboration Workloads” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shengxuan Qiu, Haochen Huang, Shuzhang Zhong, Zizhuo Fu, Zebin Yang, Meng Li
MAServeBench evaluates multi-agent language-model serving by representing collaboration rules, executing workflows to generate requests and contexts, and recording realized work in Semantic Traces for replay. It separates changes to generated work from legal changes to request release. On Qwen3-8B/910C, a fixed-work barrier reduced scheduled width from 24 to 12 and raised mean root latency 26%, in that tested setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.