Model Post-training
现代 Agent Engineering 学习路线：Model Post-training
0 子目录2 文档
文档
2Model Post-training：Agent 工程的可选高级轨道Model Post training：Agent 工程的可选高级轨道 Freshness metadata last verified : 2026 08 24 version scope : 2026 post training concepts source type : primary paper + official research stability : version sensitive 1. 先区分两条改进路径 HarTrajectory、Reward、Environment 与 Agentic RLTrajectory、Reward、Environment 与 Agentic RL Freshness metadata last verified : 2026 08 24 version scope : fast moving research area source type : primary paper stability : fast moving 数据单位 一条 trajectory 应关联 task/environme