deepseek-v4.1POST
DeepSeek open-sources DeepInfer, the inference framework serving V4.1 Flash
DeepSeek released DeepInfer, its in-house serving framework that currently runs V4.1 Flash, alongside the model; it targets high throughput and low latency to cut community compute costs and could push token prices lower.
DaDa | 🕊️
@0x99DaDaArchived 10 days agoOriginally posted 2026.10.22
DaDa | 🕊️ @0x99DaDaOriginally posted 2026.10.22
DeepSeek 开源了自家线上在用的推理框架 DeepInfer，模型和跑模型的系统都放出来了 1⃣、DeepInfer 是 DeepSeek 内部推理框架，现在跑 DeepSeek-V4.1-Flash 的就是它 2⃣、官方说主打高吞吐、低延迟，开源是为了帮社区降算力成本和能耗 3⃣、大模型开源一般只放权重，自家线上推理系统很少放出来 关键在推理这一层： 各家 API 能把价格压多低，主要看它。现在 DeepSeek 把这套系统也开源了，第三方推理商和自己部署的团队成本都可能被拉低，token 价格还有往下压的空间。 #DeepSeek #AI #开源
x.com/0x99dada/status/2113075725909872850Archived 2026.10.22 11:58