Post training deepseek with a 1k gpu Huawei cluster -- no nvidia -- feels like a serious milestone to me
"Intelligence too cheap to meter" cant arrive if there's an nvidia/tsmc bottleneck, even if its better for shareholders
Positive users hope Huawei's Nvidia-free DeepSeek training enables revolutionary open-weights local models, while negative users doubt local feasibility and criticize silicon control bottlenecks.
This is full parameter post training, no shortcuts. Probably good enough at last for a real training run
Alternate pathways to hyperscale compute are even more important if us govt can take away your models on a whim
Post training deepseek with a 1k gpu Huawei cluster -- no nvidia -- feels like a serious milestone to me
"Intelligence too cheap to meter" cant arrive if there's an nvidia/tsmc bottleneck, even if its better for shareholders
@CraigMerry Well I have bad news you will never run that locally lol
This is the only way we get a Chinese mythos imo. Just hope its open source.
@chris_j_paxton Yes indeed.
An open source alternative for that level of intuition that I can run locally would do gangbusters.
@chris_j_paxton hah. yeah definitely not today. but man. an open-weights model of that caliber locally would be revolutionary
@chris_j_paxton 1K Huawei GPUs. No Nvidia. Post-trained DeepSeek. Bottleneck isn't intelligence. It's who controls the silicon. Cheap AI requires competition, not shareholder lock-in.
@chris_j_paxton wait so the actually important part is that they trained it at all with that hardware, not just the no-nvidia headline