Back to Reading Room

LA BIBLIOTHÈQUE · READING NOTE

STaR: Bootstrapping Reasoning With Reasoning

Zelikman et al.

To Read · 想看

AIPapersReasoning

STaR, Self-Improvement, Reasoning Training, Synthetic Data

让模型自己生成推理过程,将能得到正确答案的 Reasoning Trajectory 作为训练数据继续训练模型,探索推理能力的自举与自我改进。