About
I am a Member of Technical Staff at Microsoft AI, working on AI-agent self-improvement.
I am also a Ph.D. student advised by Prof. Qiang Liu at UT Austin. Previously, I was a senior tech lead at SambaNova Systems, working on software–hardware co-design, and I earned my B.S. in Computer Science from UIUC, advised by Prof. Bo Li.
My research centers on efficient training and efficient neural architectures. I contribute actively to open-source frameworks including Hugging Face Transformers and Gradio. For more, see my research statement.
Selected Open-Source Work
C-Optim
417
Cautious Optimizers — "when it comes to optimizers, it's always better to be safe than sorry." Adopted by Hugging Face Transformers & timm. [ICLR 2026]
Python
Long-context-transformers
116
Finetuning public checkpoints on 8K-length sequences from the Pile.
Python
Super_Muon
68
Pushing the Muon optimizer further for large-model training.
Python
Online-Subspace-Descent
33
Low-rank, memory-efficient optimizer without SVD. [NeurIPS 2024]
Python
looped_nanochat
1
A looped-transformer variant of the popular nanochat.
Python
triton_kernels
1
Simplified, easy-to-understand re-implementations of Triton kernels.
Python
Recognition
Hugging Face timm adopts Cautious Optimizers
"One of the last-minute papers I added support for that delayed this release was 'Cautious Optimizers'… Consider me impressed, this boost appears more consistent than some of the new optimizers."
— Ross Wightman (@wightmanr), December 3, 2024
Writing
Selected Publications
Medical Video Generation for Disease Progression SimulationCHIL 2026 · Best Paper
Cautious Weight DecayICLR 2026