Event ·Multi-source

DeepSeek 开源推理引擎 DeepInfer，64 卡吞吐达 170 万 tok/s

DeepSeek 开源了生产环境推理引擎 DeepInfer，该引擎目前在其线上服务 V4.1 Flash。单路对话可达 325 tok/s，64 张 Nvidia Blackwell GPU 集群总吞吐约 170 万 tok/s。代码已开源，可在本地部署运行 AI 模型。

Current read

DeepSeek 把自家线上用的推理引擎 DeepInfer 开源了，单卡聊天 325 tok/s，64 张 Blackwell 跑到 170 万 tok/s，想自建推理服务的可以看看。

11 sources81° AI heatLast updated 2026/10/20 12:42:02

Evidence

11 sources

Primary source@atomic_chat_hq

2026/10/20 12:42:02View source

Related source 1kimmonismus

2026/10/20 10:26:39View source

Related source 2Teortaxes

2026/10/20 10:38:58View source

Related source 3berryxia

2026/10/20 11:34:11View source

Related source 4marktechpost

2026/10/20 14:31:59View source

Related source 5SuperTechFans

2026/10/21 00:16:33View source

Related source 6Tech in Asia

2026/10/21 05:58:27View source

Related source 7Yuchen Jin

2026/10/20 17:05:36View source

Related source 8Thomas Wolf

2026/10/18 14:34:40View source

Related source 9LMSYS Org (SGLang)

2026/10/20 00:39:06View source

Related source 10AI21 Labs

2026/10/20 11:53:16View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story