Mechanism

Samples K=100K=100 reasoning paths per math problem, verifies the final answer, clusters the correct reasoning paths by distinct intermediate equation/code signatures (discarding near-duplicate algebra paths), and trains on the deduplicated multi-path set so the student learns multiple distinct solution strategies per problem.

References

  1. Scaling Relationship on Learning Mathematical Reasoning with Large Language Models ↗

    Zheng Yuan, Hongyi Yuan, Chengpeng Li, Guanting Dong, Keming Lu, Chuanqi Tan et al. · 2023

    All 8 authors

    Zheng Yuan, Hongyi Yuan, Chengpeng Li, Guanting Dong, Keming Lu, Chuanqi Tan, Chang Zhou, Jingren Zhou