Reproduce “Agent-Conditioned Routing: Diverse Reasoning Paths Within One Frozen Model” 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
Ananth Kalyanasundaram, Ankush Kadu, Athira Gopal, Aswanth Krishnan
Agent-Conditioned Routing (ACR) uses fixed shared-and-private expert eligibility masks to constrain which experts each of four trajectories may use in a frozen mixture-of-experts model; its learned router still selects experts from each eligible set. On a screened benchmark, it yielded at least two distinct correct methods on 37.33 of 126 problems, versus 18.67 for temperature sampling at T=0.9. This result applies to selected problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.