What Is the mmdeploy 1.3.1 Documentation Website?

Updated

The mmdeploy.readthedocs.io site is the official documentation for MMDeploy 1.3.1, a model deployment toolbox for OpenMMLab models. It is the reference you use when you need to take a trained OpenMMLab model and convert, run, evaluate, quantize, or benchmark it across different inference backends and SDKs. It is documentation, not a hosted service: there is no sign-up or pricing information on the page, and nothing indicates the site itself provides compute or model hosting.

What the documentation covers

The site is organized around the deployment workflow rather than around a single product feature. The main sections listed on the landing page are:

  • Get Started — introduction, prerequisites, installation, converting a model, running inference, and evaluating a model.
  • Build — building from source, using the Docker image, building from a script, and CMake build options.
  • Run & Test — how to convert a model, how to write a config, how to evaluate a model, and how to quantize a model.
  • Useful Tools — SDK usage and SDK documentation.
  • Benchmark — supported models, benchmark results, tests on embedded devices, tests on TVM, and quantization test results.
  • OpenMMLab Codebase Support — deployment guides for individual codebases.
  • Backend Support — per-backend feature support.
  • Core Ops / Custom Ops — operator support for specific runtimes.
  • Developer Guide — architecture and extension guides.
  • Appendix and FAQ — including cross-compiling an SNPE inference server on Ubuntu 18 and frequently asked questions.

Who each section is for

If your task is… Start with
Installing MMDeploy and converting your first model Get Started (Prerequisites → Installation → Convert Model → Inference Model)
Building the toolkit yourself or in a container Build (from source, Docker image, script, CMake options)
Checking whether your model runs on a given backend Backend Support and the matching Custom Ops pages
Deploying a model from a specific OpenMMLab project The relevant codebase page (MMDetection, MMSegmentation, MMPose, MMOCR, MMAction2, MMRotate, MMDetection3d, MMagic, MMPretrain)
Comparing speed or accuracy trade-offs Benchmark, including embedded-device and TVM results
Reducing model size or latency Quantize model, plus the quantization test results
Extending MMDeploy with a new model or backend Developer Guide

Supported codebases and backends

The documentation lists deployment support for MMPretrain, MMDetection, MMSegmentation, MMagic, MMOCR, MMPose, MMDetection3d, MMRotate, and MMAction2. On the backend side, the page names ncnn, ONNX Runtime, OpenVINO, PPLNN, SNPE, TensorRT, TorchScript, RKNN, TVM, and Core ML, with separate feature-support pages for several of them. Because support differs by backend and operator, check the specific backend page and the custom-ops pages before committing to a target — a model that converts cleanly for one runtime may need custom operators for another.

Practical notes before you start

  • The site can be switched between English and Chinese using the selector in the lower-left corner of the layout, so the same content is available in both languages.
  • The documentation version is pinned at 1.3.1; if you install a different MMDeploy release, confirm the instructions still match your version.
  • The landing page does not state licensing, pricing, or login requirements for the software or the site, so treat those as things to verify from the project repository rather than assume.
  • The FAQ and Appendix are worth checking early if you hit build or cross-compilation problems, since they cover environment-specific issues such as SNPE inference server cross-compilation.

In short, use this site as your working reference for the full MMDeploy pipeline: install and build, convert and run a model, evaluate and quantize it, then verify backend and operator support before deploying to your target device.