OpenVINOOpenVINO
OpenVINO 是 Intel 的推理运行时与模型服务器：先把模型转成 IR，再在 CPU/GPU/NPU 上跑； OVMS 可暴露 OpenAI 兼容端点。
跨大类对照：Dify。
Intel's inference runtime and model server: convert the model to IR, run it on CPU/GPU/NPU, and expose OpenAI-compatible endpoints through OVMS.
适合与不适合When to use it, when not to
适合：Intel CPU/GPU/NPU、边缘部署、要把模型压到本地运行。 不适合：NVIDIA 独占优化场景、想要 GGUF 即下即跑。
Fits Intel CPU/GPU/NPU, edge deployment and squeezing models onto local hardware. Does not fit NVIDIA-only optimisation (TensorRT-LLM/vLLM are more direct) or download-and-run GGUF.
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 Intel（美国）。Intel (US).
- 国内使用Direct access
- 大陆可直连direct OK docs.openvino.ai 大陆直连未核验；GitHub/PyPI 可经镜像获取。Direct reachability from mainland not verified; GitHub/PyPI are 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）；开源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 open-source 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
把模型转换成 OpenVINO IR 后推理；支持 CPU、GPU、NPU 等 Intel 硬件。 支持模型与算子清单以官方文档为准，本站未逐条清点。
Runs models after conversion to OpenVINO IR; supports Intel CPU, GPU and NPU. The supported model and operator list follows official docs and is not enumerated here.
- 运行位置Where it runs
Runtime 库 + Model Server（OVMS）：OVMS 可暴露 OpenAI 兼容接口， 也可用 Docker 部署；本地/边缘/服务器都能跑。
Runtime library plus Model Server (OVMS): OVMS can expose an OpenAI-compatible endpoint and be deployed with Docker; it runs on local, edge and server hardware.
- 本地文件Local files
自托管，模型与 IR 文件在本地；不依赖云端账号，转换与缓存目录由部署方管理， 模型分发也由部署方负责。
Self-hosted, with models and IR files on the local machine; no cloud account dependency.
- 关机后的任务Tasks after shutdown
Runtime 是库，OVMS 是常驻服务；长任务与并发由部署方编排。
Runtime is a library while OVMS is a long-running service; long tasks and concurrency are orchestrated by the deployer.
- 工具与扩展Tools & extensions
OVMS 提供 OpenAI 兼容端点；工具调用与 MCP 支持未逐条核验。
OVMS provides OpenAI-compatible endpoints; tool-calling and MCP support 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；KV cache 与连续批处理在官方文档线。
Long context depends on model and memory configuration; KV cache and continuous batching live in official docs.
- 权限与限制Permissions & limits
本地/自托管，无账号体系；鉴权、配额与多租户隔离都由部署方在 OVMS 外层配置， 官方不提供内置租户模型。
Local/self-hosted with no account system; authentication and quotas are configured by the deployer.
- 适合什么任务Best for
适合：Intel CPU/GPU/NPU、边缘部署、要把模型压到本地运行。 不适合：NVIDIA 独占优化场景（TensorRT-LLM/vLLM 更直接）；想要 GGUF 即下即跑。
Fits Intel CPU/GPU/NPU, edge deployment and squeezing models onto local hardware. Does not fit NVIDIA-only optimisation (TensorRT-LLM/vLLM are more direct) or download-and-run GGUF.
头号误解Top misconception
- 把 OpenVINO 当成只支持 Intel CPU——官方也覆盖 GPU/NPU，但型号有矩阵Assuming OpenVINO only supports Intel CPUs — official docs also cover GPU/NPU, with a hardware matrix
- 以为模型拿来就能跑——通常要先转 IRAssuming models run as-is — they usually need conversion to IR
- 把 OVMS 当成完整 agent 平台——它只是推理服务层Treating OVMS as a full agent platform — it is only the inference service layer
未知项清单Known unknowns
- 当前支持的模型与算子清单（未逐条清点）
- OVMS 工具调用与 MCP 支持状态
- 各硬件型号的加速矩阵
- Current supported model and operator list (not enumerated here)
- OVMS tool-calling and MCP support status
- Acceleration matrix by hardware model
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 | OpenVINO · 官方文档（Runtime / 推理流水线；2026-11-03 复核官方页）OpenVINO · official docs (Runtime / inference pipeline) | https://docs.openvino.ai/ |
| docs | OpenVINO · Model Server 文档（OVMS / OpenAI 兼容端点）OpenVINO · Model Server docs (OVMS / OpenAI-compatible endpoints) | https://docs.openvino.ai/2026/model-server/ovms_docs_quick_start_guide.html |
| repo | OpenVINO · GitHub 仓库OpenVINO · GitHub repository | https://github.com/openvinotoolkit/openvino |
| changelog | OpenVINO · Releases（官方 changelog）OpenVINO · Releases (official changelog) | https://github.com/openvinotoolkit/openvino/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.
相关条目Related entries
- LemonadeLemonadeAMD 侧的本地推理服务器，与 Intel 路线对照An AMD-side local inference server, contrasting with the Intel route.
- Docker Model RunnerDocker Model Runner用容器分发与运行模型的另一条路线Another route that distributes and runs models through containers.
- DifyDify另一层 · AI 自动化与工作流Other layer · AI automation & workflows
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.