DeepSeek开源内部推理框架DeepInfer，推动大模型普惠

DeepSeek宣布将内部推理框架DeepInfer全面开源，同时开放deepseek-recipe编解码组件1。该推理框架集成了高性能算子与优化的推理引擎，支持分布式推理、推测解码、SWA Bounded Replay和结构化输出等多项功能，兼容英伟达与华为昇腾平台1。DeepSeek表示，开源这些工具的目标是优化大规模推理的吞吐与延迟，降低算力成本、减少能源消耗，从而实现大模型普惠1。

deepseek-recipe编解码组件在处理逻辑上与DeepSeek官方API实现了完整对齐1。上述推理框架和编解码组件已通过GitHub发布，项目地址分别为https://github.com/deepseek-ai/DeepInfer和https://github.com/deepseek-ai/deepseek-recipe[1](#source-1)。

DeepSeek has announced the full open-source release of DeepInfer, its internal reasoning framework, alongside the concurrent open-sourcing of the deepseek-recipe codec component 1. The initiative aims to democratize access to large language models by reducing computational costs and energy consumption 1.

DeepInfer integrates high-performance operators and an optimized inference engine, offering capabilities including distributed inference, speculative decoding, SWA Bounded Replay, and structured output generation 1. The framework is compatible with both NVIDIA and Huawei Ascend platforms 1. The deepseek-recipe component maintains full alignment with DeepSeek's official API processing logic 1. Both projects are now available on GitHub at https://github.com/deepseek-ai/DeepInfer and https://github.com/deepseek-ai/deepseek-recipe 1.

The open-source initiative is designed to optimize throughput and latency in large-scale inference workloads while lowering the barrier to entry for deploying advanced AI models 1.