Deepseek Launches DeepInfer to Cut AI Costs
Deepseek releases DeepInfer, its inference engine, as open source under the Apache 2.0 license, along with the deepseek-recipe tool to align self-hosted deployments with its API. The company claims it aims to reduce computing costs and energy consumption. This release is also presented as likely to support price competition with Western players.
The release could impact rival pricing
The open-source release of DeepInfer is seen as likely to increase pressure on the prices of Western providers such as OpenAI and Anthropic. In the context of competition between China and the United States in the field of AI, this initiative could also respond to political considerations. It is also suggested that if Western AI companies fail to meet their growth targets, it would pose a risk to the American economy and beyond. Deepseek's models are described as very affordable, and DeepInfer is presented as providing a solution to execute them even more economically. On X, Deepseek indicates its intention to help the community reduce computing costs, decrease energy consumption, and make AI more accessible.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
DeepInfer and deepseek-recipe are released under Apache 2.0
Deepseek has made DeepInfer, its internal inference framework, publicly available as open source under the Apache 2.0 license. The company also provides deepseek-recipe, a tool that replicates the token processing of Deepseek's official API to enable self-hosted configurations to operate identically. Deepseek also plans to make the code of its future models available as open source.
The framework relies on vLLM and serves V4.1-Flash
From a technical perspective, DeepInfer uses vLLM, integrates optimized kernels, leverages distributed computing across multiple processors, and employs methods that accelerate text generation. This system allows for the execution of Deepseek's V4.1-Flash model to provide rapid responses at scale. Inference here refers to the ability of a trained AI model to process queries and produce responses.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.