DeepSeek open-sources the inference engine behind its API
DeepInfer is Apache-2.0 and runs on Nvidia Blackwell-class GPUs and Huawei Ascend 950 chips; DeepSeek says one cluster serves 155 trillion tokens a day at about $38 of revenue per GPU-hour.
DeepSeek, HuaweiOpen weights & ecosystem · Compute & infrastructure · Models & capabilities