Mengsi Lyu

R&D Engineer at the Beijing Academy of Artificial Intelligence, currently serving as the head of inference compression in the AI framework team, focusing on model compression techniques, including quantization, distillation, and neural architecture search (NAS). A former employee of Baidu, she was a core developer of the open-source PaddleSlim, contributing to the design of multiple features. Previously, she led the compression efforts for the ERNIE series of models, focusing on promoting efficient model application by addressing the challenges of lossless quantization across various business needs and model sizes.