DeepSeek 开源生产推理框架 DeepInfer,支撑 DeepSeek-V4.1-Flash
DeepSeek 以 Apache 2.0 协议开源了其生产服务所用的推理框架 DeepInfer,该框架目前只支持 DeepSeek-V4.1-Flash,运行在 NVIDIA SM100 和华为昇腾 950 上。官方称单个集群每天可服务 155 万亿 token,在 131,072 上下文、64 卡解码配置下解码吞吐超过 25,000 tokens/s/GPU。此次发布仅覆盖离线推理,KV cache 系统、P/D 分离和负载均衡未包含在内,DeepSeek 表示这些组件与内部基础设施紧密耦合。
em, P/D disaggregation and load balancing are not included. What is DeepInfer? DeepInfer is a high-performance inference engine from DeepSeek for its own models. According to it’s GitHub README, it powers DeepSeek’s large-scale production services. This release covers offline inference with the core features needed at scale. It was originally built primarily on vLLM, with ideas from SGLang and TensorRT-LLM. The deployment Q&A gives production context. A single cluster serves 155 trillion tokens
匹配字段正文







