Cornfigurator is a deployment planner for generic Any-to-Any model inference serving. Given a model and workload, it explores the full spectrum of deployment strategies — from colocation to disaggregation and mixes of the two — and uses coarse-to-fine statistical evaluation to navigate the large plan space efficiently.
The current c11r planner works directly with Cornserve as its profiling and execution backend, and upcoming work will add support for vLLM-Omni and SGLang-Omni.
Source code is still under construction and will be released here soon.
For the full design and evaluation, see our paper: "Cornfigurator: Automated Planning for Any-to-Any Multimodal Model Serving".