Hacker News · 首页· ggcr·· 22 天前精选AI 评分47
DeepInfer 推理框架称单集群达 155 万亿 tokens/天
DeepInfer: High-performance inference framework for DeepSeek models
AI 导读
HN 用户 ggcr 转发 DeepInfer 推理框架的消息，称基于该框架已为 DeepSeek V4.1 Flash 提供高吞吐、低延迟的大规模推理服务，单集群服务吞吐达 155 trillion tokens/day，整体收入为 $38 per card per hour（按闲时定价），并附上知乎专栏原文链接。
正文 · 原文
A nuclear bomb
"Based on this framework, we have achieved high-throughput, low-latency large-scale inference services for DeepSeek V4.1 Flash. Currently, the service throughput of a single cluster is 155 trillion tokens/day, and the overall revenue is $38 per card per hour (at off-peak pricing)"
来源：Hacker News · 首页 · news.ycombinator.com