YUHAN SUN

y'log.

LLM、强化学习
与Agent

我是 Yuhan，写一些研究笔记和工程实践。

文章

10.162026

研究笔记11 min read

Train coordination and answering from outcomes, while a task-trained teacher guides the search tokens the student actually generates.

08.162026

研究笔记12 min read

Tools, rules, and prompts change. Training should not assume one fixed runtime.

07.132026

工程实践16 min read

从 Workflow 升级到 Agentic 架构，介绍 AgentTuning、回复模型 RLVR、端到端 Agentic RL 与多智能体协同强化学习的实践。

10.092025

研究笔记15 min read

Distill the capability, then use reinforcement learning to remove unnecessary thinking.

06.252025

工程实践15 min read

从业务背景和互动架构出发，介绍多轮意图识别、实体解析、商品检索与互动回复生成的实践。

06.202025

工程实践24 min read

从 DeepSeek-R1 蒸馏到推理路径压缩，介绍 long2short DPO、质量与长度双偏好 DP-DPO 及泛化性优化实验。

11.162023

研究笔记10 min read

Distill a prompt’s effect into LoRA, then explore the space between applying it fully and leaving it out.