Reproduce “Multi-Target Few-Shot Prompting with Diffusion Language Models” 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

Anton Xue, Alan Li, Rahul Saha

The authors present a method for jointly infilling multiple edit sites in one document with diffusion language models, which use context on both sides of masked spans. The method reweights sampled completions by their relative likelihood with demonstration outputs visible versus masked. On List Functions–Easy, LLaDA-1.5’s eight-target all-target accuracy rises from 24.3% to 44.5% under oracle lengths; guidance varies across settings.

Summary by alphaXiv.

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