Where to Find Answers to Common MMDeploy Problems

The MMDeploy 1.3.1 documentation site organizes troubleshooting resources in several places rather than a single page. Start with the FAQ section for general questions, then use the Developer Guide for model/backend extension issues, the Custom Ops pages for operator errors, and the Experimental feature pages for ONNX export and optimizer problems. The site also lets you switch between English and Chinese in the lower-left corner of the layout, which helps when a term is clearer in one language.

FAQ: general questions first

The documentation's navigation lists a FAQ entry under the appendix area. This is the first place to check for recurring setup or usage questions. If your problem is about installation, conversion, or inference basics, the FAQ is likely to cover it before you dig into developer material.

Developer Guide: when you are extending MMDeploy

The Developer Guide section targets problems that arise when you go beyond supported defaults. Its listed topics include:

• mmdeploy Architecture — understand how the pieces fit before debugging them.
• How to support new models — for models not yet covered by the supported model list.
• How to support new backends — for adding a backend that MMDeploy does not yet handle.
• How to add test units for backend ops — for validating operator behavior.
• How to test rewritten models — for checking models after rewriting.
• How to get partitioned ONNX models — for inspecting how a model was split.
• How to do regression test — for confirming changes did not break existing behavior.

These pages are the right destination when the problem is "my model/backend isn't supported" or "I need to verify a change."

Custom Ops: operator-level errors

If conversion or inference fails because of an unsupported or mismatched operator, the Custom Ops section is the relevant reference. It is organized by backend:

Use these pages to check whether an operator is supported and how it is implemented for that backend. The broader Backend Support section also lists per-backend feature support (for example, ncnn, ONNX Runtime, OpenVINO, PPLNN, SNPE, TensorRT, TorchScript, RKNN, TVM, and Core ML), which helps you confirm whether a capability exists at all before troubleshooting it.

ONNX export and optimizer issues

For problems tied to ONNX export or graph optimization, look under Experimental feature, which lists ONNX export and Optimizer. These are the pages to consult when export fails or when you want to understand what the optimizer changes.

Benchmark and test results as a sanity check

When you are unsure whether a result is a bug or expected behavior, the Benchmark section provides reference points: Supported models, Benchmark Test on embedded device, Test on TVM, and Quantization test result. Comparing your output against these can tell you whether to keep troubleshooting or accept the result.

A practical lookup order

1. Check FAQ for a known general issue.
2. Confirm backend/operator support in Backend Support and Custom Ops.
3. For extension work, go to the matching Developer Guide topic.
4. For export/optimization problems, read Experimental feature → ONNX export / Optimizer.
5. Use Benchmark results to judge whether behavior is expected.

The site also provides Index and Search pages, which are useful when you know a keyword but not its section.