DeepSeek Open-Sources DeepInfer, a High-Performance AI Inference Framework

DeepSeek has open-sourced DeepInfer, its V4.1-Flash serving stack, under the Apache 2.0 license on GitHub. This framework powers DeepSeek's large-scale production services and supports both Nvidia SM100 GPUs and Huawei Ascend 950 NPUs, aiming to make DeepSeek's recent systems papers operational.