Conversion Issues
Testing & Qualityconversion-issuesConversion Issues is an Agent Skill for testing & quality published in the openvinotoolkit/openvino repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Investigate and fix model conversion issues in OpenVINO Frontends (ONNX, PyTorch) — triage, debugging, accuracy comparison, and pre-submission verification.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/openvinotoolkit/openvino/blob/HEAD/.github/agents-prototype/skills/conversion-issues/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/conversion-issues, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Agent Skill: Investigate and Fix Frontend Conversion Issues
Goal
Diagnose and fix issues where models fail to convert to OpenVINO IR or produce incorrect inference results through an OpenVINO frontend.
Framework-Specific Workflows
Each frontend has its own detailed investigation workflow. Read the one matching the target framework:
| Frontend | Skill file | What it covers |
|---|---|---|
| ONNX | onnx.md | Triage (unsupported op / conversion bug / shape-type / opset gap), ORT baseline comparison, translator debugging, .prototxt test models, C++ GTest, pre-submission checklist |
| PyTorch | pytorch.md | Triage (unsupported op / tracing mode / inplace / normalize-step), TorchScript vs torch.export identification, layer test debugging, pre-submission checklist |
Related Skills (adding new ops)
| Frontend | Skill file | When to use |
|---|---|---|
| ONNX | add-fe-op/onnx.md | Implementing a new ONNX op translator from scratch |
| PyTorch | add-fe-op/pytorch.md | Implementing a new PyTorch op translator from scratch |
Notes
- Always verify the model works with the framework's reference runtime (ONNX Runtime / PyTorch) before investigating OpenVINO code.
- Prefer minimal, root-cause fixes over broad refactors.
- Every fix needs a test and must pass the full frontend test suite.