Check a model package description

Find JSON errors, invalid weight paths, input dimensions and profile mismatches before uploading a package.

Platform 4.0 / BOX 5.0 · Structure check

What you will get

A JSON report of errors and warnings for selected fields in model_info.json and coco_categories.json.

Requirements

  • Python 3.9 or later; no third-party libraries.
  • The extracted ZIP includes a training description and a text file in place of model weights.

Run the example

python3 check_package.py sample-package
python3 check_example.py

Expected output

"status": "structure_checked"
"errors": []
Warnings: 4 — placeholder types and profile; execution has not been checked.

Checks included

  • Required fields and types for the selected template, including positive input dimensions.
  • The weight path stays within the package directory and points to a file.
  • The default profile appears in the list, category IDs are unique, and RGB values and flags have the required types.

Scope and limitations

  • This is a preliminary check of selected fields. The complete contract depends on model_type and the target runtime.
  • The training package contains no model. Angle-bracket placeholders and weights/fixture.txt illustrate the structure; do not upload this sample package to BOX.
  • The script does not check weights, inference, additional files or profile support. A weights_path template produces a warning.

Use it with your data

  • Check a separate working copy of your extracted package.
  • Confirm supported model_type and task_type values, profiles and additional files for your build.
  • After the local check, run the platform's validation and an asset test in a test environment.

Files and source code

Related documentation