Reproduce “Asynchronous Flow Map 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
Chanhyuk Lee, Wonjung Kim, Patrick Pynadath, Jaehoon Yoo, Satoshi Hayakawa, Ruqi Zhang, Jinwoo Kim, Seunghoon Hong
The authors introduce asynchronous flow-map language models that assign tokens different denoising progress and distill this order into a single-pass predictor. At inference, one flow-map evaluation predicts the sequence. On GSM8K, the entropy-bounded student achieves 11.4% zero-shot accuracy, versus 1.6% for a synchronous student distilled from the same teacher; the result is limited to the tested setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.