Reproduce “T1: Tool-integrated Verification for Test-time Compute Scaling in Small 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
Minki Kang, Jongwon Jeong, Jaewoong Cho
The paper presents T1, a method that helps small language models verify candidate answers during test-time search. T1 uses tools such as code execution to check calculations before a reward model ranks candidates. On MATH500, adding ToolV raises weighted Best-of-N accuracy for a 1B model at 64 candidates from 47.4% to 53.2%, but tool checks can reject correct answers.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.