PeopleofSuperintelligence

Builder

Yuanzhong Xu

Yuanzhong Xu is a Google engineer whose parallelism systems train giant models, including Gemini. In 2016 he completed a doctorate with Emmett Witchel at the University of Texas at Austin on platform-level protection for interacting mobile apps, joined Facebook, and shared an OSDI best paper for Ryoan, a distributed sandbox for untrusted computation on secret data. He coauthored GShard, which scaled a translation model beyond 600 billion parameters, and in December 2021 presented GSPMD, an automatic parallelization system for the XLA compiler. In July 2023 he was a senior staff software engineer at Google Brain, now Google DeepMind, with GSPMD powering training and serving for PaLM, LaMDA, and Gemini. The December 2023 Gemini report named him a co-lead for codebase and parallelism, and as of October 2026 he is at Google DeepMind on Gemini.

Works
42

Publication numbers from OpenAlex, as of 17 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2016

    A doctorate, then Facebook

    In 2016 he completes a doctorate with Emmett Witchel. The dissertation is Platform-level Protection for Interacting Mobile Apps, and he joins Facebook.

  2. 2016

    An OSDI best paper

    In 2016, with Tyler Hunt, Zhiting Zhu, Simon Peter, and Emmett Witchel at the University of Texas at Austin, he shares an OSDI best paper for Ryoan, a distributed sandbox for untrusted computation on secret data.

  3. 2021

    GShard at ICLR 2021

    He coauthors GShard for ICLR 2021. The system scales a multilingual translation Transformer with a Sparsely-Gated Mixture-of-Experts beyond 600 billion parameters, and the model trains on 2048 TPU v3 accelerators in 4 days.

  4. December 2021

    Presenting GSPMD

    In December 2021 he and Yanping Huang, software engineers on the Google Research Brain team, present GSPMD, an automatic parallelization system based on the XLA compiler. The system is applied to GShard-M4, LaMDA, BigSSL, ViT, and MetNet-2, and it is integrated into TensorFlow and JAX.

  5. July 2023

    Senior staff at DeepMind

    In July 2023 he is a senior staff software engineer at Google Brain, now Google DeepMind, with a focus on automated parallelism for large models. GSPMD powers training and serving for PaLM, PaLM2, Gemini, Parti, USM, LaMDA, and Bard. He holds a PhD from UT Austin and a BS from SJTU.

  6. December 2023

    Gemini codebase co-lead

    In December 2023 the Gemini team at Google submits a report on a family of multimodal models for image, audio, video, and text. The report names him co-lead for codebase and parallelism.

  7. As of October 2026

    Google DeepMind on Gemini

    As of October 2026, he is at Google DeepMind on Gemini.

Sources and credits

Photo: We have no published photo of Yuanzhong Xu yet. Send us a free-licensed photo.

  1. Emmett Witchel Alumni cs.utexas.edu. Checked 16 October 2026.
  2. Ryoan: A Distributed Sandbox for Untrusted Computation on Secret Data | USENIX usenix.org. Checked 16 October 2026.
  3. GShard: Scaling Giant Models With Conditional Computation and Automatic Sharding research.google. Checked 16 October 2026.
  4. General and Scalable Parallelization for Neural Networks research.google. Checked 16 October 2026.
  5. 对话大模型 ｜ 硅谷创业与投资沙龙7月聚会 – 上海交通大学硅谷校友会 sjtu-sv.org. Checked 16 October 2026.
  6. Gemini: A Family of Highly Capable Multimodal Models arxiv.org. Checked 16 October 2026.
  7. [2312.11805] Gemini: A Family of Highly Capable Multimodal Models arxiv.org. Checked 16 October 2026.
  8. ukoxyz (Yuanzhong Xu) · GitHub github.com. Checked 16 October 2026.