TMTPOST — DeepSeek has decided to fully open-source its internal high-performance inference framework DeepInfer. The framework underpins the company’s large-scale production services for DeepSeek V4.1 Flash, delivering high throughput and low latency. Future inference code and optimizations for new models will continue to be updated in the repository. DeepInfer integrates optimized kernels and supports distributed inference features, and is compatible with both NVIDIA and Huawei Ascend platforms.