Reproduce “Reducing Batch Makespan for LLM Agent Workloads through Inference-Environment Coordination” 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

Wei Feng, Meiling Chen, Boxiao Chen, Lu Liu, Bin Zi, Libin Ye, Xingyu Lu, Lei Xu and 1 more

The authors present a system for reducing batch makespan in agentic LLM development workloads. It coordinates inference-cache allocation and trajectory-aware routing with separate sample admission and environment execution limits, while CAAS prioritizes queued jobs using observed progress. In MCP-Atlas evaluation, Opt-CAAS reduces makespan from 466.1 to 163.0 minutes, with scores of 65.5 and 65.6; this combined-configuration result does not isolate CAAS.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.