Reproduce “Gumbel Distillation for Parallel Text Generation” 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
Chi Zhang, Xixi Hu, Bo Liu, Qiang Liu
The paper introduces Gumbel Distillation, which conditions parallel language models on teacher-associated Gumbel noise to predict target tokens in parallel. For its main experiments, the authors extract noise from existing text, not teacher-generated sequences. On OpenWebText, MDLM improves MAUVE from 0.217 to 0.282 and lowers generative perplexity from 38.34 to 34.33 versus its undistilled version; these quality gains do not measure wall-clock speed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.