概念
reasoning chain-of-thought
创建2026-04-15
更新2026-04-15
阅读量级1 分钟
概念导读

STaR (Self-Taught Reasoner) is a method where language models generate their own reasoning traces (chain-of-thought) to solve problems, then fine-tune on their own successful reasoning. This creates a self-improvement loop that enhances reasoning capabilities without external teacher models. STaR is related to on-policy-self-distillation in that both approac...

  1. Related
  2. References

STaR (Self-Taught Reasoner)

STaR (Self-Taught Reasoner) is a method where language models generate their own reasoning traces (chain-of-thought) to solve problems, then fine-tune on their own successful reasoning. This creates a self-improvement loop that enhances reasoning capabilities without external teacher models. STaR is related to on-policy-self-distillation in that both approaches leverage the model's own outputs for training, though STaR focuses specifically on reasoning trace generation.

References

  • Zelikman et al. (2022). "STaR: Bootstrapping Reasoning With Reasoning." NeurIPS.