Try it with 2 GPUs: sglang serve --model-path tencent/Hy-Nomas --num-gpus 2 --… · @GenAI_is_real · TwKit

@GenAI_is_real

Try it with 2 GPUs: sglang serve --model-path tencent/Hy-Nomas --num-gpus 2 --sp-degree 2 --port 30000 For now, Director and Generator are served separately. We're looking forward to working with the Hunyuan team on unified end-to-end serving, including coordinated KV cache scheduling and inference across both models. Huge thanks to the Tencent Hunyuan team for open-sourcing HY-Nomas, and to everyone who helped bring it into SGLang!!!

曝光 99 · 评论 1 · 点赞 4 · 书签 0