Reproduce “Weave of Formal Thought” 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
Alexandre Bouayad
The paper presents WoFT, a code-training method that interleaves AST-derived non-terminals with code and learns which formal tokens to retain, alongside a Tree-sitter-based constrained-decoding engine. On StarCoder2-3B, WoFT lowers per-text-token cross-entropy versus text-only training in all ten tested languages, averaging 11.61% lower loss; this does not establish syntax validity or functional correctness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.