Cheng Li

Cheng Li

Senior Researcher

Microsoft

I am senior software engineer at Databricks GenAI. My work has focused on understanding and optimizing inference/training of Deep Learning (DL) models, particularly on Transformers (LLMs). Before that, I was a senior researcher at Microsoft. At Microsoft, I worked on improving the performance/usability of transformer models in production (e.g. GitHub Copilot, DALL·E-2, etc.), building systematic profiling/optimization stacks for DL, and integrating SOTA system technologies into Microsoft DeepSpeed, an open-source DL optimization software suite that enables unprecedented scale and speed for training and inference.

I recently developed and open sourced llm-analysis: Latency and Memory Analysis of Transformer Models for Training and Inference, check it out!

Interests

  • Deep Learning and Transformers (LLMs)
  • System Optimization and Engineering for Deep Learning
  • GPU and Parallel Computing

Education

  • PhD in Computer Science, 2020

    University of Illinois Urbana-Champaign

  • MS in Computer Science and Engineering, 2015

    University of Michigan

  • BS in Computer Engineering, 2013

    University of Michigan

  • BS in Electrical Engineering, 2013

    Shanghai Jiao Tong University

Latest