DeepSeek Open-Sources DeepInfer Inference Framework, Cuts Compute Cost on Nvidia and Ascend
On October 20 DeepSeek fully open-sourced its internal inference framework DeepInfer, sharing large-scale cloud-service deployment practice for DeepSeek V4.1 Flash. The f…
On October 20 DeepSeek fully open-sourced its internal inference framework DeepInfer, sharing large-scale cloud-service deployment practice for DeepSeek V4.1 Flash. The framework supports distributed inference, speculative decoding and structured output, and runs on both Nvidia and Huawei Ascend platforms. DeepSeek disclosed a single-cluster throughput of 155 trillion tokens per day and revenue of 38 USD per card per hour at trough pricing, saying the goal is to lower compute cost and energy use so intelligence reaches more people at lower price - a step toward true large-model inclusiveness.
Source: Southern Metropolis Daily / NetEase, Oct 21, 2026 | 2026-10-20
延伸阅读：
- Nine Departments Issue Plan to Turn Brain-Computer Interfaces, Embodied AI Into Market Applications
- DeepSeek Open-Sources DeepInfer Inference Framework, Cuts Compute Cost on Nvidia and Ascend
- BAAI Releases FlagOS 2.2 Open AI-Compute Stack Covering 21 Chip Makers
- 32nd China Yiwu International Small Commodities Fair Opens With 2,770 Exhibitors
- Chinese AI Model Weekly Token Calls Top U.S. for 25th Straight Week; Demand Volatility Noted
- Shandong Unveils 'Qilu Seven Stars': Seven Vertical Industry Large Models at AI Industry Conference
- China Releases First Batch of Standards for AI in the Steel Industry
- Smart Products Become Yiwu's New Foreign-Trade Growth Driver as U.S. Buyers Add Orders After Tariff Cuts
- Over 100 South American Buyers Arrive in Groups for the 32nd Yiwu Fair
- Yiwu Turns Into a 'World Live-Streaming Room' as Foreign Hosts Broadcast Brands Globally
