Xianwei Zhuang

About Me

I am a second year graduate student majoring in computer science at Peking University, mainly focusing on foundational large model, including the pre/post-training and RLHF of large vision models (VLM).

Research Interests

• Multimodal Large Language Model: Unified Understanding and Generation, SFT, RLHF, Reasoning [QwenVLo, VARGPT]
• Large Language Model: Long Context [Doubao]

Experience

Educational Experience

• Peking University: Pursuing a master’s degree starting from 2023
• Zhejiang University: Obtained a Bachelor’s degree from 2019 to 2023

Intern

• Qwen Team, Foundation Model, Tongyi: Core contributor in the 0-to-1 development of Qwen VLo, participating in data synthesis, coding, pre/post-training, and RL.
• ByteDance: Focus on studying long context of LLM and long-term memory of Doubao

Some honors

• Received the National Scholarship at Peking University in 2024
• Received Zhejiang Province Outstanding Graduate and Zhejiang University First Class Scholarship at Zhejiang University

News

• [Jun. 2025] Our QwenVLo preview version can available for you to experience at QwenVLo.
• [Apr. 2025] Open source training code and models of VARGPT-v1.1 at github and Huggingface. Total GitHub stars exceed ~600.
• [Feb. 2025] Our works VASparse is accepted for CVPR 2025. Work is continuously being updated at github .
• [Jan. 2025] Our code, model and paper about unified understanding and generation VARGPT-v1 has been released.
• [Jan. 2025] Our work on CoT Distillation initially completed in Feb. 2024 is accepted for ICLR 2025.
• [Dec. 2024] Our paper and code about Hallucination Mitigation for MLLMs VASparse has been released.

Publications: VLM Pre-/Post-training, RLHF, Unified Modeling

Publications: VL, CoT Distillation, NLU

Powered by Jekyll and Minimal Light theme.