AI People Wiki

Builders & open source

Tri Dao

Assistant Professor of Computer Science, leading the Dao AI Lab, Princeton University

Tri Dao is an Assistant Professor of Computer Science at Princeton University, where Dao leads the Dao AI Lab, and a co-founder and the Chief Scientist of Together AI; Dao's site describes research on efficient training and inference through hardware-aware algorithms. Dao is first author of the FlashAttention paper and sole author of FlashAttention-2, whose official implementation lists support for consumer GPUs such as the RTX 3090 and RTX 4090. With Albert Gu, Dao co-authored the Mamba and Mamba-2 papers, and the Mamba repository publishes pretrained checkpoints from 130M to 2.8B parameters on Hugging Face under the Apache 2.0 license.12345678

At a glance

Current roles
Assistant Professor of Computer Science, leading the Dao AI Lab +1 more
Sector
University
Education
Princeton, Stanford
Work listed
2
Honors
6
Sources cited
16
Profile reviewed

Career

Dao completed a PhD in computer science at Stanford University from 2016 to 2023, advised by Christopher Ré and Stefano Ermon, with a dissertation on hardware-aware algorithms for efficient machine learning.10

Dao co-founded Together AI in July 2023 and serves as its Chief Scientist, and joined Princeton University as an assistant professor of computer science in September 2024, where Dao leads the Dao AI Lab. Dao's research covers machine learning and systems, focused on hardware-aware algorithms and sequence models with long-range memory.101

Research

Dao's FlashAttention (2022) computes exact attention with an IO-aware algorithm that reduces reads and writes to GPU memory, and FlashAttention-2 (2023) improved its parallelism and work partitioning. FlashAttention-3 (2024) targeted asynchrony and low precision on newer GPUs, and FlashAttention-4 appeared at MLSys 2026. Dao's CV notes that FlashAttention has been integrated into frameworks including PyTorch, JAX, Hugging Face Transformers, vLLM and SGLang.2311104

With Albert Gu, Dao co-authored Mamba (2023), a selective state space model for sequence modelling, and Mamba-2 (2024), which connects state space models and attention through structured state space duality; later work includes Mamba-3. Earlier, Dao was first author of Monarch (2022), on expressive structured matrices for efficient training.5671012

Work and contributions

  • Mamba and Mamba-2

    Co-author with Albert Gu of the Mamba (2023) and Mamba-2 (2024) papers, and listed in the repository's AUTHORS file. The Apache-2.0 repository publishes pretrained Mamba and Mamba-2 checkpoints from 130M to 2.8B parameters on Hugging Face, with model cards under the Apache 2.0 license, and includes a script that downloads a checkpoint from the Hub and generates text.56713148

  • FlashAttention

    First author of the FlashAttention paper (2022) and sole author of FlashAttention-2 (2023). The BSD-3-Clause repository, whose AUTHORS file lists Dao, provides the official implementation; its README lists support for Ampere, Ada, and Hopper GPUs including the RTX 3090 and RTX 4090, and a ROCm version for AMD GPUs.2341516

Timeline

  1. Began a PhD in computer science at Stanford University, advised by Christopher Ré and Stefano Ermon.10
  2. Posted "Monarch, Expressive Structured Matrices for Efficient and Accurate Training" as first author.12
  3. Posted the FlashAttention paper as first author.2
  4. Co-founded Together AI as Chief Scientist and released FlashAttention-2 as sole author.103
  5. Co-authored Mamba, linear-time sequence modeling with selective state spaces, with Albert Gu.5
  6. Co-authored Mamba-2 and the structured state space duality framework with Albert Gu.6
  7. Co-authored FlashAttention-3, on fast attention with asynchrony and low precision.11
  8. Joined Princeton University as an assistant professor of computer science.10
  9. Co-authored FlashAttention-4, presented at MLSys 2026.10

Honors and recognition

  • ICML Outstanding Paper runner-up · International Conference on Machine Learning (ICML), 2022101
  • Stanford Open Source Software Prize (inaugural, for FlashAttention) · Stanford University, 202410
  • COLM Outstanding Paper · Conference on Language Modeling (COLM), 2024101
  • Schmidt Sciences AI2050 Fellowship · Schmidt Sciences, 2025101
  • Google Research Scholar · Google, 2025101
  • MLSys Best Paper Honorable Mention · Conference on Machine Learning and Systems (MLSys), 20261

Where they work

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

16 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. Tri Dao — homepageTri Dao · Official page · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  2. FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness (arXiv 2205.14135)arXiv · Paper · published May 27, 2022 · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  3. FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning (arXiv 2307.08691)arXiv · Paper · published Jul 17, 2023 · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  4. Dao-AILab/flash-attention README (installation and GPU support)Dao-AILab (GitHub repository) · Repository · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  5. Mamba: Linear-Time Sequence Modeling with Selective State Spaces (arXiv 2312.00752)arXiv · Paper · published Dec 1, 2023 · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  6. Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality (arXiv 2405.21060)arXiv · Paper · published May 31, 2024 · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  7. state-spaces/mamba README (pretrained models and inference) and license (Apache-2.0)state-spaces (GitHub repository) · Repository · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  8. state-spaces/mamba-2.8b model card (license apache-2.0)state-spaces (Hugging Face) · Model card · published Dec 3, 2023 · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  9. tridao (Tri Dao) on GitHubGitHub · Official page · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  10. Tri Dao — Curriculum Vitae (updated 01/2026)Tri Dao · Official page · published Jan 2026 · accessed Oct 21, 2026
  11. FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision (arXiv 2407.08608)arXiv · Paper · published Jul 11, 2024 · accessed Oct 21, 2026
  12. Monarch: Expressive Structured Matrices for Efficient and Accurate Training (arXiv 2204.00595)arXiv · Paper · published Apr 1, 2022 · accessed Oct 21, 2026
  13. state-spaces/mamba AUTHORSstate-spaces (GitHub repository) · Repository · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  14. state-spaces on Hugging Face (pretrained Mamba and Mamba-2 checkpoints)state-spaces (Hugging Face) · Official page · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  15. Dao-AILab/flash-attention AUTHORSDao-AILab (GitHub repository) · Repository · accessed Oct 21, 2026 · reviewed Oct 2, 2026
  16. Dao-AILab/flash-attention LICENSE (BSD 3-Clause)Dao-AILab (GitHub repository) · License · accessed Oct 21, 2026 · reviewed Oct 2, 2026