Reproduce “From Solver Trajectories to Teaching Trajectories: Cognition-Aligned Reasoning Distillation” 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
Yashuo Luo, Qianren Mao, Ziqi Qin, Junnan Liu, Haitao Li, Qili Zhang, Yuan Gao, Zhijun Chen and 1 more
CARD reuses existing teacher trajectories, rewriting rationales with detected failed steps while preserving answers and leaving other examples unchanged. On Qwen2.5-7B, it scores 54.22 Math Avg and 71.07 General Avg, versus 51.80 and 56.34 for Vanilla SFT. These macro-averages support gains in this mathematical-distillation setup, evaluated only with students up to 7B parameters.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.