Reproduce “Workload-Static Optical Circuit Switching For Scalable LLM Inference Decode” 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
Maximilian Thürmer, Nicolas Scotto di Perto, Eloise Stein, Natanijel Vasic, Mikaeil Bagherli, Zhenguo Wu, Benjamin Klenk, Holger Fröning and 1 more
The authors jointly choose model parallelization, logical network topology, and collective schedules for a deployment-static optical circuit fabric serving recurring decode communication. They model rather than demonstrate hardware, comparing a GPT-OSS-5.3-like workload at batch size 1 and 5 million tokens of context. All-OCS reaches 1,277 tokens/s/user versus 775 for all-EPS, a 65% gain, in a non-blocking OCS versus 4:1-oversubscribed EPS comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.