Welcome to vLLM office hours. These bi-weekly sessions are your chance to stay current with the vLLM ecosystem, ask questions, and hear directly from contributors and power users. This week’s special topic: vLLM-triton-backend: How to get SOTA performance on NVIDIA and AMD with just Triton We kick off with the latest vLLM project update from core committer Michael Goin, followed by a deep dive with Burkhard Ringlein from IBM Research on how to achieve state-of-the-art performance across NVIDIA and AMD GPUs using Triton. Burkhard walks through IBM’s new Triton backend for vLLM, which eliminates the need for hand-written CUDA or HIP kernels, delivering top-tier performance and portability with a single Triton-only code base through advanced autotuning and optimized kernel design. Session slides: https://docs.google.com/presentation/d/1pku9S5VnMaGYs3xp0wMy1s9qB15xWQWCfF-Z1-8fQPc/ Time stamps 00:00 Welcome, agenda, upcoming vLLM meetups 03:41 About vLLM 06:37 vLLM project and community update 08:17 Accelerating LLMs with NVFP4 quantization 23:50 Deep dive: vLLM-triton-backend 56:20 Live Q&A, feedback, and wrap-up Want to join the discussion live on Google Meet? Get a calendar invite by filling out this form: https://red.ht/office-hours

 478      11