A Shenzhen team completed full-parameter post-training of DeepSeek-V4-Pro (1.6 trillion parameters) on a Huawei Ascend 910C cluster. MFU exceeded 30%, and the training ran 1,500+ steps without interruption.
Two things to keep in mind. This is post-training, not pre-training. The compute requirements are an order of magnitude smaller. And 30% MFU is below what NVIDIA clusters typically achieve (40-50%+).
Still, the fact that a third-party team (not Huawei itself) pulled this off on domestic chips is a meaningful data point for the Ascend ecosystem's maturity.
@AndrewCurran_ and in fact, they are right after the mess we all saw here in the security sector, what they did will be done in bio too, too dangerous.
he is telling: IA bot traffic is more than humans now.
if you aren't preparing hour SEO and your pages to it... you REKT.
at this point is ai agents that will visit your parge from now.
Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history. https://t.co/2zX5bHdhsa