vLLM 官方博客T1·Infra
78 pts
vLLM-Omni has brought MiniMax's open-weight H3-Image to NVIDIA Vera Rubin, achieving a 0.87-second median for native 4096×4096 generation on four GPUs, with a throughput configuration reaching 6,115 images per hour. A single checkpoint handles both text-to-image and editing in six sampling steps, producing native high-resolution output without an upscaler and supporting up to 16 reference images.