Hao AI Lab· @haoailab · X·· 4 小时前AI 评分43
最多4台DGX Spark跑全栈
AI 导读
(3/7) 整套技术栈最多可在 4 台 NVIDIA DGX Spark 上运行 我们的全本地部署基于 NVIDIA DGX Spark 构建。单台 Spark 的 128 GB 统一内存可同时容纳导演智能体(通过 @vllm_project 提供服务)和视频模型。深度集成 FastVideo 服务后,你最多可将四台 Spark 互联扩展,以获得更强算力与速度。
正文 · 原文
(3/7) Runs the entire stack on up to 4 NVIDIA DGX Sparks
Our fully local setup is built upon NVIDIA DGX Spark. One Spark is capable of holding the directing agent (served on @vllm_project ) and the video model in its 128 GB of unified memory at the same time. With deep integration of FastVideo serving, you can scale up to four Sparks together for more compute and speed.
来源:Hao AI Lab · x.com
