Papers: 'Exploring the Impact of Model Scaling on Parameter-efficient Tuning Methods' accepted by EMNLP 2023; 'ChatDev' accepted by ACL 2024; 'AgentVerse' and 'ChatEval' accepted by ICLR 2024. Project contributions: Created APRIL and released it on GitHub; created rlsys dockerhub to support RL training framework on AMD MI-series GPUs; integrated AMD ROCm support into slime and verl.
Research Experience
Postdoctoral Researcher at CMU/MBZUAI, hosted by Eric Xing. Now a research scientist at AMD GenAI team.
Education
Ph.D. in Department of Computer Science and Technology at Tsinghua University from 2019 to 2023, advised by Zhiyuan Liu and part of THUNLP Lab hosted by Maosong Sun.
Background
LLM researcher/engineer, with research interests in foundation models, particularly focusing on pre-training/post-training frameworks and training efficiency optimization. Currently working as a research scientist at AMD's GenAI team.