Agent Algorithm Engineer at Tencent IEG · M.Sc., Shanghai Jiao Tong University
I work on LLM post-training, reinforcement learning, and agent systems.
First author · Accepted to EMNLP 2026 Main Conference
GAC adaptively balances supervised fine-tuning and reinforcement learning using a noise-aware mixing controller. The project includes a reference implementation, a public model checkpoint, and a versioned evaluation profile.
I was responsible for GAC's method design, theoretical derivations, training implementation, experiments, and open-source maintenance.
Paper · Code · Project page · Model · Evaluation profile
See the publication index for paper links, version-specific BibTeX citations, and structured metadata.
- LLM post-training and SFT–RL optimization
- Reinforcement learning for language models
- Agent systems and evaluation
胡玥麟，上海交通大学硕士，现任腾讯 IEG 中台 Agent 算法工程师。研究与工程方向为大语言模型后训练、强化学习与智能体系统。第一作者论文 GAC 被 EMNLP 2026 主会接收。本人负责 GAC 的方法设计、理论推导、训练实现、实验和开源维护。