DeepSeek V4 Flash 0731 (Q4) now reaches 1,328 tok/s prefill and ~29 tok/s decode on one RTX PRO 6000

I've been working on speeding up DeepSeek-V4-Flash-0731 in Krasis and have now got the long-prompt prefill quite a bit faster on a single RTX PRO 6000 96GB. These are timing-disabled internal Krasis results using INT4 experts. They aren't HTTP round-trip speeds: Prompt size Prompt Processing ab

中文全文翻译正在补齐，可先查看下方英文原文。

I've been working on speeding up DeepSeek-V4-Flash-0731 in Krasis and have now got the long-prompt prefill quite a bit faster on a single RTX PRO 6000 96GB. These are timing-disabled internal Krasis results using INT4 experts. They aren't HTTP round-trip speeds: Prompt size Prompt Processing ab