DeepSeek 开源了自家生产级推理框架 DeepInfer，还原生支持华为昇腾 950。
This is a dev post classified by AGK Intelligence as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
DeepSeek 开源了自家生产级推理框架 DeepInfer，还原生支持华为昇腾 950。 项目采用 Apache-2.0 协议，可商用、修改。它源自 DeepSeek 大规模推理服务，不过目前开放的是离线推理版本，并非完整线上部署方案。 几个重点： 硬件支持：英伟达 SM100 GPU、华为昇腾 950 NPU 模型支持：目前仅支持 DeepSeek-V4.1-Flash 推理加速：通过 DSpark 草稿 token 和置信度验证提升解码效率 底层算子：集成 FlashMLA、DeepGEMM、DeepEP 等内核 结构化输出：支持正则、JSON Schema 和 EBNF 约束 另外，批次不变推理和确定性采样目前仅支持英伟达平台。 需要注意：KV Cache、Prefill/Decode 分离及副本间负载均衡等线上服务组件暂未包含。项目早期主要基于 vLLM 构建，也参考了 SGLang 和 TensorRT-LLM。 🔗 https://github.com/deepseek-ai/DeepInfer
Posted by lumxss (34.5k followers) 1 h ago · 2 likes · 171 views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: deepinfer.
More dev work like this
- Voice Remixing for Gemini 3.8 Flash TTS is now live in @GoogleAIStudio and the Gemini API. — @Google
- Omarchy Replay (Remember RewindAI?) records one display, reads its text locally, and… — @rblalock
- they did it. — @fanofaliens
- Atmanity has launched Atmee v2, turning one image and 10 seconds of audio into an AI… — @wallstengine
- kimi k3.1 pro weights are officially out — @notjazii
- what if a 50 cm boulder chased my Microduck? would it learn how to avoid it? — @witcheer
- At the movies watching Clayface and decided to keep working lol — @Da7_Tech
- Is that skill worth the tokens? 🤔 — @rseroter
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.3k posts from 5k X accounts over the last 21 days, 9.3k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-04 17:16 UTC. Full method.