AI GitHub
ModelScan for inspecting serialized model files
Look for suspicious operations in supported serialized model artifacts before loading them.
Key Takeaways
- ModelScan examines model files for indicators such as risky serialization operations.
- Put the scanner before any step that loads model artifacts.
- Scan known-safe fixtures and controlled examples from the project's tests.
- A static scan covers recognized patterns and formats.
How it works
ModelScan examines model files for indicators such as risky serialization operations. It fits before model loading in an intake or CI process.
What you need
Put the scanner before any step that loads model artifacts. Keep the file hash and findings together so the same file can be reviewed without running it.
What to try first
Scan known-safe fixtures and controlled examples from the project's tests. Confirm that unknown formats and scanner errors lead to a reviewable result.
Add an unrecognized file type to an intake test. Observe whether the surrounding workflow routes it for review or mistakes a skipped scan for a pass.
Where to look for failures
A static scan covers recognized patterns and formats. Keep model provenance and isolation requirements in place for new artifacts.
Related Franklin pages
Separate artifact screening from execution. Test how your intake pipeline handles a flagged file or unknown format before passing a model to any inference runtime.
- Inspect pickle opcodes for risky imports and related indicators. PickleScan for screening Python pickle artifacts
- Execute exported ONNX models with execution providers suited to the target device. ONNX Runtime for portable inference
For broader questions about the risks of deploying AI systems, see Franklin’s glossary. AI safety
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗