Browse publications on Google Scholar (top-right) ↗
Resume (English only)
Academic Achievements
August 2025, a paper accepted by EMNLP 2025; our latest work on LLM preference optimization (DuPO) is now available as a preprint.
July 2025, released two key projects: (1) Seed-X: 7B-parameter multilingual translation LLM with open-sourced models and demos; (2) Seed LiveInterpret 2.0: End-to-end simultaneous speech-to-speech translation system with 3-second latency (70% reduction from prior solutions) and voice cloning.
2022, Excellent Doctoral Paper Award, Jiangsu Association of Artificial Intelligence (JSAI).
Technical Reports: 'Seed LiveInterpret 2.0' and 'Seed-X: Building Strong Multilingual Translation LLM with 7B Parameters'.
Publications/Preprints include: 'DuPO: Enabling Reliable LLM Self-Verification via Dual Preference Optimization', 'EDT: Improving Large Language Models by Entropy-based Dynamic Temperature Sampling', etc.
Research Experience
Interned at ByteDance AI Lab, researching deep generative models (such as non-autoregressive text generation, latent variable modeling). Currently, a Research Scientist at ByteDance Seed.
Education
Earned a Ph.D. in March 2022 from the Natural Language Processing Group of Nanjing University, co-supervised by Prof. Shujian Huang and Prof. Jiajun Chen. During doctoral studies, interned at ByteDance AI Lab, mentored by Prof. Zhou Hao and Prof. Lei Li, where I researched deep generative modeling (e.g., non-autoregressive text generation, latent variable modeling).
Background
Currently a Research Scientist at ByteDance Seed, specializing in Large Language Model (LLM), AI for Science (AI4S), and Natural Language Processing (NLP) research.
Miscellany
Long-term hiring! Now join us to build cutting-edge AI systems. Open positions include: (1) Top Seed Program (rolling recruitment for PhD candidates & recent graduates, internships & full-time roles available), reach out via baoyu.001@bytedance.com or (2) direct applications via job listings. Feel free to inquire about role details anytime.