Reproduce “VLM Fine-Tuning for End-to-End Combinatorial Optimization” 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

Qingsong Yan, Xia Jiang, Yaoxin Wu, Wen Song, Lu Zhang, Yingjie Zhou

The paper presents a shared vision-language solver for seven combinatorial optimization tasks that receives exact text and an instance-derived image. It fine-tunes Qwen3-VL to generate solutions, then uses feasibility and objective-quality rewards to select candidates. In CVRP, the VLM reports a 3.58% gap versus 4.53% for the text-only solver, with 98% versus 100% feasibility; the comparison does not isolate images.

Summary by alphaXiv.

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