Reproduce “Rate-Resolved Text Evaluation with Masked 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
Yiqi Liu, Kun Zhao, Chenghua Lin, Nafise Sadat Moosavi
The authors use a masked diffusion model to evaluate text by masking fractions of tokens, reconstructing them, and retaining losses by rate. They use conditional token revision frequency to localize errors. On LLaDA, a selected rate band raises SummEval relevance Spearman correlation to 0.2795, versus 0.2610 with a collapsed diffusion scalar; localization varies by metric, and rate meanings require task-specific validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.