Announcement
@saltyph reports DeepSeek v4.1 Flash at roughly 2,500 tokens per second on Vera Rubin
RadixArk says it’s collaborating to bring Baseten’s low-latency optimizations to SGLang and the broader community.
TLDR
The speed claim comes with specifics: @saltyph reports roughly 2,500 tokens per second for DeepSeek v4.1 Flash on Vera Rubin at batch size 1, TP4, and 1.9 milliseconds per step. The post claims performance up to 1.7× faster than its reported GB300 results. RadixArk says it’s collaborating to bring Baseten’s optimizations into SGLang.
Combined views
1.8K
3 Sources, first seen ago

Comments