← Runners

Docker Model RunnerDocker Model Runner

DockerDocker

DockerDocker 本地推理Local inference OCIOCI OpenAI兼容OpenAI-compatible partial lifecycle: active 核验Checked 2026-11-03

Docker Model Runner 把模型当成 OCI 工件：从 Docker Hub、OCI registry 或 Hugging Face 拉取，在本机通过 OpenAI/Ollama 兼容 API 提供服务。

Models as OCI artefacts: pull from Docker Hub, any OCI registry or Hugging Face and serve locally through OpenAI- or Ollama-compatible APIs.

适合与不适合When to use it, when not to

适合：已用 Docker 的团队、想把模型分发纳入容器流水线、要本地 OpenAI 兼容端点。不适合：不想装 Docker、要 GPU 极限优化、要成品聊天界面。

Fits teams already on Docker who want model distribution inside container pipelines and a local OpenAI-compatible endpoint. Does not fit people avoiding Docker, extreme GPU optimisation, or a ready-made chat UI.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-11-03

出品属地Origin: 国外出品foreign-made　Docker（美国）。Docker (US).
国内使用Direct access: 大陆可直连direct OK　docs.docker.com 大陆直连未核验；Docker Hub 可经镜像获取。Direct reachability from mainland not verified; Docker Hub is available through mirrors.
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-11-03 检索）；开发工具未面向大陆公众提供服务，备案不适用。Not in the CAC public registry (searched 2026-11-03); the developer tool does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 从 Docker Hub、任意 OCI 兼容 registry 或 Hugging Face 拉取模型并运行；支持 LLM 与其他 AI 模型，具体模型清单以官方文档为准。
Pulls models from Docker Hub, any OCI-compatible registry or Hugging Face and runs them; supports LLMs and other AI models, with the exact list following official docs.
运行位置Where it runs: 集成在 Docker Desktop 与 Docker Engine：本地拉起模型，通过 OpenAI 与 Ollama 兼容 API 提供服务（官方文档）。
Integrated into Docker Desktop and Docker Engine: starts models locally and serves them through OpenAI- and Ollama-compatible APIs (official docs).
本地文件Local files: 模型作为 OCI 工件存储与分发；提示与数据在本机处理，不需要 API key。
Models are stored and distributed as OCI artefacts; prompts and data are processed locally with no API key needed.
关机后的任务Tasks after shutdown: 常驻本地服务，可被 Docker 生态内的应用调用；没有「关机后续跑」的官方声明。
A long-running local service callable by apps in the Docker ecosystem; no official statement of continuing after shutdown.
工具与扩展Tools & extensions: OpenAI/Ollama 兼容端点；Docker Agent 可自动发现已拉取模型。工具调用与 MCP 支持未逐条核验。
OpenAI/Ollama-compatible endpoints; Docker Agent can discover already-pulled models. Tool-calling and MCP not verified item by item.
上下文与记忆Context & memory: 长上下文按模型与内存配置；官方提供低内存、长文档等配置预设，单次请求的上下文上限由所选模型决定。
Long context depends on model and memory configuration; official presets cover low-memory and long-document setups.
权限与限制Permissions & limits: 本地 Docker 环境内运行；权限由 Docker 与宿主系统决定。
Runs inside the local Docker environment; permissions come from Docker and the host system.
适合什么任务Best for: 适合：已用 Docker 的团队、想把模型分发纳入容器流水线、要本地 OpenAI 兼容端点。不适合：不想装 Docker、要 GPU 极限优化、要成品聊天界面。
Fits teams already on Docker who want model distribution inside container pipelines and a local OpenAI-compatible endpoint. Does not fit people avoiding Docker, extreme GPU optimisation, or a ready-made chat UI.

头号误解Top misconception

把 DMR 当成云端模型服务——它在本机跑，数据不出本机Treating DMR as a cloud model service — it runs locally and data stays on the machine
以为所有镜像都能直接跑模型——模型要按 DMR 的 OCI 工件格式分发Assuming any image can run a model — models are distributed as DMR OCI artefacts
忽略 Docker Desktop 许可——商用前读 Docker 条款Ignoring Docker Desktop licensing — read Docker terms before commercial use

未知项清单Known unknowns

当前支持的模型格式与完整清单
工具调用与 MCP 支持状态
GPU 加速在各平台的支持矩阵

Current supported model formats and full list
Tool-calling and MCP support status
GPU acceleration matrix across platforms

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-11-03，内容更新日content updated on 2026-11-03。

类型Type	名称Name	链接Link
docs	Docker · Model Runner 官方文档（OCI/HF 拉取、OpenAI/Ollama 兼容 API；2026-11-03 复核官方页）Docker · Model Runner official docs (OCI/HF pulls, OpenAI/Ollama-compatible APIs)	https://docs.docker.com/ai/model-runner/
docs	Docker · Get started with DMR（启用、拉取、运行、发布）Docker · Get started with DMR (enable, pull, run, publish)	https://docs.docker.com/ai/model-runner/get-started/
docs	Docker · DMR configuration（配置预设）Docker · DMR configuration (presets)	https://docs.docker.com/ai/model-runner/configuration/
changelog	Docker · model-runner GitHub Releases（官方 changelog）Docker · model-runner GitHub Releases (official changelog)	https://github.com/docker/model-runner/releases

实测记录Measured runs

本档未跑吞吐与显存实测；结论只覆盖运行时能力与许可。

This site has not completed throughput or VRAM tests; the conclusions cover runtime capability and licensing only.