Biography
Hi, there! This is Zhisheng Ye. I am currently a machine learning systems researcher at Applied Machine Learning at Bytedance, where I work on building efficient and practical systems for emerging recommendation models and LLM workloads.
I received my Ph.D. in Institute of Networking and Energy-efficient Computing (NEEC) at Peking University in 2024, under the joint supervision of Prof. Yingwei Luo, the director of NEEC, and Prof. Xiaolin Wang. Previously, I received a B.S. degree in Computer Science and Technology from the School of Electronics Engineering and Computer Science (EECS) at Peking University, China, in 2019.
My research focuses on building efficient systems for ML and emerging LLM workloads across the entire stack, spanning training frameworks, resource scheduling, and GPU/HPC optimization, as a former member of PKUSC. I also received mentorship from Prof. Tianwei Zhang of NTU and Peng Sun, and had strong collaborations with his students, including Wei Gao, Qinghao Hu, Meng Zhang, and Qiaoling Chen.
Download my CV.
- AI Infrastructure for LLMs
- Machine Learning Systems
- Resource Management
-
Ph.D. in Computer Architecture, 2024
Peking University
-
BSc in Computer Science and Technology, 2019
Peking University
Experience
- Large scale model (e.g., LLM, MoE) training infrastructure optimization.
- Deeply involved in the development of InternLM.
- Supercomputing cluster scheduling and optimization for deep learning training workloads in Sensetime Research (now SenseCore).
- Design and implementation of a fair scheduler for DLT jobs as first author.
- Contributed to development of OpenI-Octopus, an open-sourced scheduler for deep learning training workloads based on Kubernetes.
- Safe GPU sharing and efficient migration mechanisms on Kubernetes.
- Monitoring and logging systems.
- Participated in analyzing, compiling, profiling, optimizing, and improving parallelizability of general HPC tasks.
- First Price (Team), ASC19 Student Supercomputer Challenge