SemiAnalysis_· @SemiAnalysis_ · X·· 2 天前AI 评分
Inferact 与 NVIDIA 联合优化 DeepSeek v4.1 Flash 推理
Inferact 与 NVIDIA 联合优化 DeepSeek v4.1 Flash 推理，已提交 200 tok/s/user p90 交互性 SLO 曲线，未来数日将扩展至 150 和 100 tok/s/user。
Currently, Inferact & NVIDIA have only submitted up to 200 tok/s/user p90 interactivity SLO, but in the coming days, they will also be extending the curve to 150 tok/s/user and 100 tok/s/user, which will be up to 10-20x better than any other submission and make all the other submission curves look like flat lines because of how good this co-designed technique is.
DeepSeek v4.1 Flash work was done by Juntian, Yifan, the Inferact team, and the NVIDIA team! Awesome work!
Link to dashboard (3/3) inferencex.semianalysis.com/…
来源：SemiAnalysis_ · x.com