ديب سيك تفتح مصدر البرنامج اللي تشغّل فيه نماذجها، عشان تشغيل الذكاء الاصطناعي يصير أرخصDeepSeek open-sources the software it uses to run its models, aiming to make AI cheaper to run

China's DeepSeek has published DeepInfer, the software it uses to run its models for users, as open source on GitHub under the Apache 2.0 licence, The Next Web reported.

What does running a model mean? It is the work that happens every time someone sends the model a request to answer. The same model can cost much more or much less to run depending on the software serving it. DeepSeek says it is sharing the tool to help developers lower compute costs and energy use and make AI more affordable.

For now the tool supports only DeepSeek-V4.1-Flash and runs on Nvidia's SM100 chips and Huawei's Ascend 950 chips. It includes a technique called speculative decoding, which drafts several pieces of the answer ahead and checks them in one pass. By the afternoon of the release day, the repository had more than 8,000 stars on GitHub.

But it is not a complete service on its own, and leaves out parts including a cache system and load balancing across machines. The published performance figures are DeepSeek's own, and there is no independent benchmarking yet, Bloomberg reported. The release came a day after the US ordered civilian federal agencies to remove DeepSeek products from their systems.