Liang Sheng is magnificent! DeepSeek's inference framework is fully open-sourced, disclosing daily average throughput and gross profit—the real OpenAI has arrived.

What happened: DeepSeek open-sourced DeepInfer, its large-scale cloud inference framework, and deepseek-recipe, a production-grade token codec component. Why it matters: DeepSeek says V4.1 Flash's high-throughput, low-latency serving runs on this framework, and it works on both Nvidia and Huawei Ascend platforms. The team noted KVCache management and load balancing code is not fully open-sourced yet due to coupling with internal infrastructure, and future inference code will keep landing in DeepInfer. Coverage: 3 outlets in 24h

DeepSeek open-sourced DeepInfer, its large-scale cloud inference framework, and deepseek-recipe, a production-grade token codec component.

DeepSeek says V4.1 Flash's high-throughput, low-latency serving runs on this framework, and it works on both Nvidia and Huawei Ascend platforms. The team noted KVCache management and load balancing code is not fully open-sourced yet due to coupling with internal infrastructure, and future inference code will keep landing in DeepInfer.