DeepSeek open-sources internal inference framework DeepInfer

Open-sourcing an inference framework that already runs a production model is a different move from publishing weights: it exposes the serving layer where cost and latency are decided, not just the model itself. The commitment to keep updating the repository for future models matters more than the initial release, since it signals the framework is meant to track DeepSeek's roadmap rather than stand as a one-off artifact. Whether outside teams adopt it, and whether the promised updates keep pace with new model releases, is the open question.