> [!NOTE] > The IBM Granite 4.2 disclosure files described below are currently being prepared and will be published in this repository soon. Please check back for updates.
Overview This folder contains model and data metadata disclosures in machine-readable JSON format for IBM Granite 4.2 models. IBM is an industry leader in AI transparency and governance, as recognized by IBM's top ranking on the Stanford Foundation Model Transparency Index (FMTI). These disclosures take our commitment to transparency a step further, moving away from free-form, unstructured technical disclosures towards a structured approach to releasing metadata in machine-readable format to inform risk-based controls at runtime.
How to use the Disclosures These disclosures can be parsed and extracted for various purposes, including to: aid in the filing of reporting requirements (e.g., the Model Documentation Form for the GP AI Code of Practice, EU AI Act) and/or trigger risk-based controls (e.g., linking safety policies at runtime to data metadata disclosures for data used in the data mix for model training).
What Comes Next? This is the first step towards a broader AI Bill of Materials (AIBOM) for Granite models. The objective of these disclosures is to bring structured model metadata disclosures closer to the application of risk-based controls at runtime. Our goal is to allow software engineers and other developers to apply safety policies at runtime that are informed by the risks.
License All Granite 4.2 disclosures are distributed under CDLA Permissive V2 licenses.
Would You Like to Provide Feedback? Please let us know your comments about our disclosures or provide feedback to the team on metadata and/or other reporting metrics that are most important for your industry or use case. You can do so by posting comments on our GitHub discussions page.