DeepSeek has open-sourced DeepInfer, its high-performance AI inference infrastructure, under the Apache 2.0 license.
DeepSeek has open-sourced DeepInfer, its high-performance AI inference infrastructure, under the Apache 2.0 license.
Built to run DeepSeek-V4.1-Flash, the framework combines optimized GPU kernels, parallel execution, and asynchronous scheduling to improve large-scale model inference. The initial release supports NVIDIA and Huawei accelerators and provides offline inference, with additional components required for a complete production service.