In the coming week, developers can expect five repos in the form of “daily drops” from Chinese LLM maker DeepSeek. The team will share more expertise around the building blocks of its own online services.
DeepSeek diverged from established AI players such as OpenAI and Anthropic by open-source its advanced AI models. At least, it involved the LLMs themselves, so they could be run by any user with enough computing power. A scientific paper additionally revealed all sorts of details about DeepSeek’s methods. Still, some details are missing. For example, what data was used to train DeepSeek-V3 and DeepSeek-R1? And what are some of the details around the AI infrastructure behind the online chatbot? That last question will be answered next week.
Also read: LLM hijackers target DeepSeek API
AI infrastructure
The initiative is called DeepSeek Open Infra. A new GitHub repository is shared daily to accompany DeepSeek’s claims with actual code.
The promise for the five repos is that they are the “humble building blocks” for DeepSeek’s online service. “Documented, deployed and battle-tested,” is how the AI maker puts it on GitHub. And, to get ahead of cynical predictions: “no vaporware, just programming code that moved our tiny moonshot forward.”
The repos are shared with the generous Creative Commons (CC) 1.0 license. This allows users to copy, modify, distribute and run the code, including for commercial purposes. No permission is required for this either. However, a link to the original code and acknowledgement is required to comply with the CC 1.0 license.