About Me

Hi, welcome to my homepage. I am Chen Liang (梁辰), a Member of Technical Staff at Microsoft AI. Before Microsoft, I completed my Ph.D. at Georgia Tech, advised by Prof. Tuo Zhao.

I currently work on post-training for coding agents across Microsoft and OpenAI model families. Previously, I worked on code-domain mid-training.

More broadly, my research focuses on efficient and generalizable LLM training, guided by a “less is more” principle: efficiency can improve model quality by revealing the essential learning signals. I study this through data-efficient training, stable optimization, model compression, parameter-efficient adaptation, and scaling in RL.

Recent News

  • Jun. 2026. We introduced MAI-Code-1-Flash, a 5B-active-parameter SWE agent for GitHub Copilot in VS Code and CLI.
  • Apr. 2026. The NorMuon optimizer was accepted to ICML 2026 as a Spotlight. It ranked #1 on the NanGPT speedrun leaderboard in Oct. 2025 [record, post].
  • May. 2025. We introduced GPT-4o Copilot, a GPT-4o-based code-completion model for VS Code.
  • Mar. 2025. We released Phi-mini-MoE and Phi-tiny-MoE, compact MoE models designed for edge-device settings with 2.4B / 1.1B activated parameters with 1M+ monthly downloads. See SlimMoE for details.
  • Aug. 2024. We introduced Phi-3.5-MoE, a 16×3.8B MoE model pre-trained from scratch. See GRIN-MoE and Phi-3 Technical Report for details.
  • Feb. 2024. I completed my Ph.D. thesis, "On Parameter Efficiency of Neural Language Models" [thesis, talk], and joined Microsoft as a Senior Researcher.
  • Jan. 2024. LoftQ, a LoRA-aware quantization framework, was accepted to ICLR 2024 as an Oral [talk, blog].

Selected Publications (Google Scholar, *Equal Contribution)

Shuffle the Context: RoPE-Perturbed Self-Distillation for Long-Context Adaptation

Zichong Li, Chen Liang, Liliang Ren, Tuo Zhao, Yelong Shen, Weizhu Chen

The 43th International Conference on Machine Learning (ICML), 2026
Arxiv
NorMuon: Making Muon More Efficient and Scalable

Zichong Li*, Liming Liu*, Chen Liang, Weizhu Chen, Tuo Zhao

The 43th International Conference on Machine Learning (ICML Spotlight), 2026
Arxiv NanoGPT Speedrun Leaderboard Github
SlimMoE: Structured Compression of Large MoE Models via Expert Slimming and Distillation

Zichong Li, Chen Liang, Zixuan Zhang, Ilgee Hong, Young Jin Kim, Weizhu Chen, Tuo Zhao

The Second Conference on Language Modeling (COLM), 2025
Arxiv Phi-tiny-MoE-instruct Phi-mini-MoE-instruct Phi-3 Technical Report
Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling

Liliang Ren, Yang Liu, Yadong Lu, Yelong Shen, Chen Liang, Weizhu Chen

The 13th International Conference on Learning Representations (ICLR), 2025
Arxiv Github
LoftQ: LoRA-Fine-Tuning-Aware Quantization for Large Language Models

Yixiao Li*, Yifan Yu*, Chen Liang, Pengcheng He, Nikos Karampatziakis, Weizhu Chen, Tuo Zhao

The 12th International Conference on Learning Representations (ICLR Oral), 2024
Arxiv Blog Ckpt Github
Module-wise Adaptive Distillation for Multimodality Foundation Models

Chen Liang, Jiahui Yu, Ming-Hsuan Yang, Matthew Brown, Yin Cui, Tuo Zhao, Boqing Gong, Tianyi Zhou

The 37th Conference on Neural Information Processing Systems (NeurIPS), 2023
Arxiv
Less is More: Task-aware Layer-wise Distillation for Language Model Compression

Chen Liang, Simiao Zuo, Qingru Zhang, Pengcheng He, Weizhu Chen, Tuo Zhao

The 40th International Conference on Machine Learning (ICML), 2023
Arxiv Github Ckpt
HomoDistil: Homotopic Task-Agnostic Distillation of Pre-trained Transformers

Chen Liang, Haoming Jiang, Zheng Li, Xianfeng Tang, Bing Yin, Tuo Zhao

The 11th International Conference on Learning Representations (ICLR), 2023
Arxiv
PLATON: Pruning Large Transformer Models with Upper Confidence Bound of Weight Importance

Qingru Zhang, Simiao Zuo, Chen Liang, Alexander Bukharin, Pengcheng He, Weizhu Chen, Tuo Zhao

The 39th International Conference on Machine Learning (ICML), 2022
Arxiv Github
No Parameters Left Behind: Sensitivity Guided Adaptive Learning Rate for Training Large Transformer Models

Chen Liang, Haoming Jiang, Simiao Zuo, Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen, Tuo Zhao

The 10th International Conference on Learning Representations (ICLR), 2022
Arxiv Github
Super Tickets in Pre-Trained Language Models: From Model Compression to Improving Generalization

Chen Liang, Simiao Zuo, Minshuo Chen, Haoming Jiang, Xiaodong Liu, Pengcheng He, Tuo Zhao, Weizhu Chen

The 59th Annual Conference of the Association for Computational Linguistics (ACL), 2021
Arxiv Github

Experience

Senior Researcher, Microsoft AI, Feb. 2024 – Present
Research Intern, Microsoft Azure AI, Google Research, and Amazon, summers 2021–2023
SDE Intern, NVIDIA, May 2018 – Aug. 2018

Education

Ph.D. in Machine Learning, Georgia Institute of Technology, School of Industrial and Systems Engineering, Dec. 2023

B.S. in Electrical Engineering, University of Southern California, Department of Electrical and Computer Engineering, May 2018

Service & Teaching

Area Chair: NeurIPS 2025-2026
Reviewer: NeurIPS 2021–2024, ICML 2021–2025, ICLR 2021–2023, COLM 2024, EMNLP 2021–2022, ACL/NAACL 2021–2022
Teaching Assistant, ISyE 3030 Basic Statistical Methods, Georgia Tech, Fall 2020
Teaching Assistant, ISyE 3770 Statistics & Applications, Georgia Tech, Summer 2020
Teaching Assistant, CSE 6140 Algorithms, Georgia Tech, Fall 2019
Course Producer, EE 364 Introduction to Probability & Statistics, USC, Fall 2017

Talks

Dec. 2023. LoftQ: LoRA-Fine-Tuning-Aware Quantization, NeurIPS Third Workshop on Efficient Natural Language and Speech Processing
Sep. 2023. On Parameter Efficiency of Neural Language Models, Allen Institute for AI