Yulong Ao
Head of AI Framework Development at BAAI
Yulong Ao is the Head of AI Framework Development at BAAI, a Postdoctoral Fellow at Peking University, and holds a Ph.D. from the Chinese Academy of Sciences. He currently leads the development of the large-model training and inference framework FlagScale and the unified communication library FlagCX, both key components of the open-source unified AI software stack FlagOS. He pioneered industry-ready solutions for heterogeneous large-model training and inference across diverse compute architectures. His long-term research focuses on distributed systems and performance optimization in artificial intelligence, high-performance computing (HPC), and scientific computing. Previously, he worked at Huawei and Baidu, contributing to core technologies in large-model systems. In 2016, he jointly received China’s first ACM Gordon Bell Award. He has published over ten papers in top international conferences and journals, holds multiple domestic and international patents, and has participated in the formulation of national and international standards for operator interfaces and communication libraries.
Topic
Unified Computing Power, Unleashing Intelligence: The Evolution of FlagScale within the FlagOS Ecosystem
Abstract: The AIGC wave has accelerated the diversification of AI computing power, leading to the emergence of various chips. However, this fragmentation of computing resources has resulted in complex model deployment, challenging optimization, and rising costs. To address these issues, **FlagOS** has developed a unified AI infrastructure designed for multi-chip ecosystems. This report will provide an in-depth introduction to its two core components: **FlagScale** and **FlagCX**. The former enables integrated training and inference optimization through automatic tuning, multi-backend support, and heterogeneous mixed training and inference mechanisms; the latter establishes unified communication standards to support efficient cross-architecture collaboration. Together, they demonstrate how AI computing systems are evolving from fragmentation toward convergence.
