Jul 04, 2024 Out latest work Lazarus for resilient and elastic training of Mixture-of-Experts models is available. Check it out!
Jun 29, 2024 Out latest work VcLLM is now online. VcLLM utilizes video codecs to compress various types of tensors in LLM training and inference. Check it out!
May 04, 2024 Out work on re-architecting collective as a system service, MCCS, got accepted by SIGCOMM 2024.
May 22, 2023 I start an internship at Networking Research Group in Microsoft Research, working with Wei Bai.