The goal of Oak ML Transparency is to generate non-forgeable claims about machine learning models by evaluating them against an evaluation script in a secure environment and embedding the result of the execution as a JSON object in a claim generated and signed by the secure execution environment.

To the core of this design is a runner that runs the script and generates the claim. For more details please see the linked documentation.

Currently, the runner can run an evaluation script that is a python script, and accepts the following input flags:

• --model: input path in the local file system for loading the model
• --output: output path where the script stores the result of the evaluation

The result of the evaluation can be anything, but we recommend using a JSON document, since the generated claim itself is a JSON object.

For details about the claim, see the claim format proposed by our transparent-release project.