{"rewrite":{"id":"r_f1c16df9a97f478ba6045f53","clusterId":"c_884b28fdb5e53a5e08e12805","slug":"deepseek-opens-its-production-inference-system-deepinfer","model":"deepseek-v4-1-flash","headline":"DeepSeek Opens Its Production Inference System DeepInfer","summary":"DeepSeek published DeepInfer, the inference system its own AI services use, with core functionality packaged for outside deployment. It is built on vLLM with ideas from SGLang and TensorRT LLM, runs DeepSeek-V4.1-Flash, and supports the DSpark acceleration technology, but a KV cache system is not implemented. DeepSeek also released deepseek-recipe, which converts requests and outputs between third-party APIs and the official DeepSeek API format.","whyItMatters":"Open-sourcing the stack the company actually runs shifts the practical question for third parties from whether they can reproduce DeepSeek's API behavior to whether they can do it without the KV cache the release leaves out.","webCardHtml":"\u003cp\u003eDeepInfer carries the Apache License Version 2.0 and sits in DeepSeek\u0026#39;s GitHub organization. The release notes a gap: no KV cache system is implemented, so what ships is the core functionality rather than the full serving stack.\u003c/p\u003e\u003cp\u003eThe companion repository, deepseek-recipe, comes under the MIT License. Its conversion tools let third-party API providers reproduce behavior equivalent to the official DeepSeek API, which matters for anyone reselling or proxying the models.\u003c/p\u003e","blueskyPost":"DeepSeek has open-sourced DeepInfer, the inference system its own services run on. It builds on vLLM with ideas from SGLang and TensorRT LLM, supports DeepSeek-V4.1-Flash and DSpark, and ships without a KV cache. License: Apache 2.0. A second repo, deepseek-recipe, is MIT.","twitterPost":"DeepSeek open-sources DeepInfer, the inference system used in its production services. Built on vLLM with ideas from SGLang and TensorRT LLM. Runs DeepSeek-V4.1-Flash, supports DSpark, no KV cache implemented. Apache 2.0. deepseek-recipe also out under MIT.","threadsPost":null,"newsletterBlurb":"DeepSeek published DeepInfer, the inference system its own AI services use, under Apache License 2.0, built on vLLM with ideas from SGLang and TensorRT LLM. It runs DeepSeek-V4.1-Flash and supports DSpark acceleration, but ships without a KV cache system. A second repository, deepseek-recipe, is out under the MIT License and can convert requests and outputs so third-party API providers match the official DeepSeek API format.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261021-deepseek-deepinfer-recipe/\",\"title\":\"DeepSeek releases AI inference system \\\"DeepInfer,\\\" an open-source version of the system actually used in DeepSeek's production services\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4664,"outputTokens":649,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1792563730,"createdAt":"2026-10-21T06:17:52.000Z","publishedAt":"2026-10-21T06:21:08.000Z","updatedAt":"2026-10-21T06:21:08.000Z"},"cluster":{"id":"c_884b28fdb5e53a5e08e12805","canonicalTitle":"DeepSeekがAI推論システム「DeepInfer」を公開、DeepSeekの本番サービスで実際に使われているシステムのオープンソース版","representativeArticleId":"a_f3d74292043b9f69d7ed8837","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"DeepInfer\"],\"studios\":[\"DeepSeek\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-10-21T01:53:00.000Z","lastSeenAt":"2026-10-21T01:53:00.000Z","updatedAt":"2026-10-21T06:21:07.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261021-deepseek-deepinfer-recipe/","title":"DeepSeekがAI推論システム「DeepInfer」を公開、DeepSeekの本番サービスで実際に使われているシステムのオープンソース版"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["DeepInfer"],"studios":["DeepSeek"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}