Kaichao You

Kaichao You's site lists a research record in deep learning and transfer learning, much of it with Mingsheng Long and Jianmin Wang. Papers include "How Does Learning Rate Decay Help Modern Neural Networks?" (2019, with Michael I. Jordan), LogME (ICML 2021), a practical method for assessing pre-trained models for transfer learning, and Efficient ConvBN Blocks for Transfer Learning and Beyond (ICLR 2024), which the site says was integrated into MMCV and into PyTorch starting with version 2.2.18910

You is first author of depyf, a tool that decompiles and exposes the workings of the PyTorch compiler for machine learning researchers, published in the Journal of Machine Learning Research in 2025. The site describes depyf as a PyTorch ecosystem project that was highlighted in the opening keynote of PyTorch Conference 2024.561

You is a core maintainer of vLLM, the open-source engine for serving large language models. The project's committers document names You as a project lead for distributed execution, compilation and community and as an owner of the plugin interface, and the governance document lists You among both the core maintainers, who serve as the Technical Steering Committee, and the lead maintainers. You also co-authored Jenga (SOSP 2025), on memory management for serving language models with heterogeneous components.43111

In January 2026, creators and core maintainers of vLLM announced Inferact, a company whose stated mission is to grow vLLM as an inference engine and make inference cheaper and faster. You is a co-founder and the Chief Scientist of Inferact.712