Hi, I'm Chenxi Ma, a third-year undergraduate student currently interning at UW Systems Lab under the guidance of Baris Kasikci.
My interests lie in LLM inference optimization and GPU systems. I explore how model capabilities connect with the systems that execute them.
My technical interests include inference runtimes, GPU scheduling, mixture-of-experts models, and communication and computation in distributed inference. I'm also interested in co-design across model architectures, agent harnesses, hardware, and infrastructure.
I'm a contributor to vLLM, working on MoE expert parallelism, dual-batch overlap (DBO), and NIXL-based dispatch in Model Runner V2 (MRV2).
This page brings together my projects and research interests. Technical articles and experiment notes live on my separate blog.
Exploring efficient model execution and reliable reasoning and decision-making.
An inference runtime, from request to next token.
A readable re-implementation of DeepInfer's core design, running the real DeepSeek-V4.1-Flash on 4 × B200. Every optimization is recorded in a GitHub issue with its baseline, profile, correctness tests and measured result. Prefill throughput is up about 4× so far, and still 8.4× behind the official engine.
MoE expert parallelism for vLLM.
Fused Triton kernels, 2× NCCL all-to-all, and CUDA Graph decode on B200.
Selected contributions to vLLM, focusing on MoE expert parallelism, dual-batch overlap (DBO), and NIXL dispatch. Thanks to everyone who helped along the way.
Report the first recorded microbatch failure so that follow-on errors do not hide the original problem.
Preserve expert routing at each microbatch's token offset, fixing overwritten routing results.
Size NIXL dispatch from synchronized DP / SP execution sizes to reduce unnecessary work for small batches.
