LLM Reasoning Framework。Chain-of-Thought CoT (Wei et al)・Tree-of-Thoughts ToT (Yao et al)・Self-Consistency CoT-SC・Reflexion・ReAct (Reasoning+Acting)・Self-Discover (Google)・DSPy 2.6 Stanford (Auto Prompt Optimize)・Process Reward Model PRM800K (OpenAI)・Outcome Reward Model ORM・Verify Step-by-Step・LATS Language Agent Tree Search・¥0 OSS、2026年DSPy+PRM RLHF Reasoning訓練主流。