Reproduce “A Survey on Continuous Diffusion and Flow 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
Sophia Tang, Chanhyuk Lee, Floor Eijkelboom, Jinwoo Kim
The authors survey continuous language models for parallel, few-step text generation, explaining how shared sequence flows couple token positions and how learned flow maps make finite jumps. They contrast this with independent marginal sampling and discuss RePlaid’s training-compute gap: 20× with self-conditioning, down from Plaid’s 64×. Scaling matched discrete diffusion over the evaluated compute range; practical efficiency remains open.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.