Reproduce “Training-Free Character-Length Control in Summarization 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
Steffen Castle, Nils Feldhus, Christopher Ebert, Leonhard Hennig, Sebastian Möller
DELTA controls character length in summaries by resizing a masked buffer during diffusion generation. It estimates length from committed text and predictions for masked positions, then inserts or removes masks; this requires no inference-time training. On arXiv summaries with LLaDA-8B-Instruct, mean absolute length error is 2.8 characters and 98.4% fall within 10 characters, but targets are reference-summary lengths, not display slots.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.