Pingzhi Tang
Undergraduate Student at Peking University (Tong Class).
Advancing Efficient LLM Architectures.
Occasional observer of the world's quiet moments.
Undergraduate Student at Peking University (Tong Class).
Advancing Efficient LLM Architectures.
Occasional observer of the world's quiet moments.
I am Pingzhi Tang, an undergraduate student in the General Artificial Intelligence Experimental Program (Tong Class) at Peking University.
With a major GPA of 3.92/4.00, my research focuses on enhancing the performance and efficiency of large-scale models, specifically in model architecture design and efficient inference.
Beyond the code, I am a photography enthusiast and film lover. I find that the intuition used to optimize a neural network often resonates with the process of composing a frame—both are pursuits of finding structure within complexity.
Tong Class.
Major GPA: 3.92/4.00.
* See CV for full list of awards.
Advisor: Prof. Muhan Zhang.
Working on efficient inference, model architecture, PEFT and LLM reasoning.
Addressed KV cache overhead in MLA tensor parallelism. Achieved 2x inference acceleration.
Fanxu Meng*, Pingzhi Tang*, Zengwei Yao, Xing Sun, Muhan Zhang
Proposed a method to transform standard multi-head attention into multi-head latent attention, achieving up to 11x inference acceleration with negligible loss.
Yiding Wang, Fanxu Meng, Xuefeng Zhang, Fan Jiang, Pingzhi Tang, Muhan Zhang
Invented a distributed PEFT method for LLMs that increases update flexibility by assigning different weight components to each GPU.
Fanxu Meng, Pingzhi Tang, Fan Jiang, Muhan Zhang
Developed a novel large model pruning and fine-tuning method using cross-layer singular value decomposition.
For a complete list of publications, please visit my Google Scholar or CV.
Coming soon.
