DeepSeek Open-Sources DeepInfer Inference Framework and Token Codec

Key Info

DeepSeek open-sourced DeepInfer, its in-house inference framework used to serve DeepSeek-V4.1-Flash at scale with high throughput and low latency, and also released deepseek-recipe, a production token encoding and decoding component aligned with the official DeepSeek API's processing logic.

Highlights

• DeepInfer is the production inference stack behind DeepSeek-V4.1-Flash, aimed at high throughput and low latency
• DeepSeek says sharing the stack is meant to help the community lower compute costs and energy use
• deepseek-recipe brings the same token encoding/decoding behavior as the official API to community deployments
• The release spans open models, open kernels and open inference, targeting both private deployments and public clouds
• Code is available for community exploration and contribution