Show HN: Optimized vLLM for Blackwell (SM_120) – 59T/S on DeepSeek-R1

github.com/informatico-madrid

2 points

malka666

5 months ago

vLLM for Blackwell (SM_120) & Kernel 6.14. Optimized for RTX 5090. 59 t/s on DeepSeek-R1 via FlashInfer