Yuxuan Tian 田宇轩
I am a graduate student at the School of Software and Microelectronics, Peking University, advised by Prof. Tong Yang. I received my B.S. in Computer Science and Technology from Peking University in 2025.
My research focuses on LLM pre-training, post-training, and efficient inference. I am currently interested in improving mathematical and reasoning capabilities through high-quality data and training recipes, and reducing inference cost through system-level optimization.
Interests
数学&推理预训练
Pre-training for mathematical and reasoning capabilities through high-quality data and training pipelines.
数学&推理后训练
Post-training for mathematical reasoning with high-quality data, reasoning SFT, RL and Model Merge.
高效推理
Efficient LLM inference, KV cache compression, and serving-time system optimization.
Education
Scholarships & Awards
| Year | Scholarship / Award |
|---|---|
| 2021-2022 | 北京大学一等奖学金、北京大学三好学生 |
| 2022-2023 | 北京大学三等奖学金、北京大学三好学生 |
| 2023-2024 | 北京大学冈松奖学金、北京大学三好学生 |
| 2025 | 北京大学优秀毕业生 |
News
- 2026小红书 Hi Lab-dots预训练 实习；数学与推理预训练pipeline建设
- 2025360智脑 实习；大模型数学推理与Agentic Coding能力后训练研究
- 2025就读北京大学软件与微电子学院硕士研究生
- 2024字节跳动 Seed-Infra 实习；KV Cache 压缩与推理优化
Projects
北京市自然科学基金
2023 - 2025
北京市自然科学基金本科生“启研”计划。担任项目负责人，开展高速数据流测量相关研究；相关成果发表于 ACM SIGKDD 和 ACM IMC。
华为“难题揭榜”火花奖
2024
参与华为“难题揭榜”第八十八期-难题2“MoE训练的动态专家负载预测技术”，研究 MoE 训练过程中的动态专家负载预测，获得第97期火花奖；相关成果形成论文 Prediction Is All MoE Needs。
TinyR1-32B
2025
负责 TinyR1-32B 模型的数学&推理领域后训练，围绕高质量数学数据、reasoning SFT 与 branch-merge distillation 提升模型能力，使模型数学能力达到同尺寸 SOTA 水平。Hugging Face
Research
- AAAI 2026, Oral
- arXiv 2025
- ACM SIGKDD 2025
- ACM IMC 2025
- arXiv 2024