2025-04-08 03:15:18
LRM论文选读:START: Self-taught Reasoner with Tools 这篇文章介绍了一种构建特定任务、私域Large Reasoning Model的有效方法,主要方法是微调,而且没有使用reinforcement learning,但却有非常类似reinforcement learning with human feedback的特征。 关键词有二: Test-time scaling Rejection sampling fine-tuning(反思微调)