@nvidia's Lee Nau and Akhil Goel share what it took to bring SGLang to Vera Rubin, find performance bottlenecks, and tune kernels and attention for large MoE models.
NVIDIA engineers explain tuning SGLang for Vera Rubin and large MoE models
AINVIDIA engineers Lee Nau and Akhil Goel describe what it took to bring SGLang to Vera Rubin. They cover finding performance bottlenecks and tuning kernels and attention for large MoE models.
Source details
SGLang · x.comPublished · added here
Original title: @nvidia's Lee Nau and Akhil Goel share what it took to bring SGLang to Vera Rubin, find performance bottlenecks, and tune kernels and att...