As part of our commitment to Trustworthy AI, NVIDIA developed Model Card++ to uplevel AI model documentation originally proposed by data scientists at Google.

Model Card is helping us steward and democratize AI responsibly. Model Card documentation gives model consumers (namely enterprise developers as well as nontechnical influencers like customers, investors, users, and policymakers) information about key metadata fields in a clear and uniform manner. Model Cards provide more than expected performance data; they clarify intended domain, applications, use cases; restrictions around use; dataset characterizations, and known technical limitations and risk. Model cards also help developers more quickly adopt, deploy, and build state of the art software. Producing model card documentation is a state-of-the-art best practice for advancing AI transparently and trustworthily in the software development lifecycle.

Following an engineering pilot and longitudinal model card market research, NVIDIA affirmed we could uplevel our model card documentation to provide Model Card++, which provides a common software quality expectation for our work and the data that underpins our artificial intelligence (AI) models. Our Model Card++s are everywhere our models are made available: NGC Catalog & Private Registry, NVAIE, GitHub, and partner software cloud platforms.

Please use the templates provided in the repo as you see fit per the Creative Commons Zero (CCO) License, adopting as you see fit. If you have recommendations or questions, please file an issue. Stay tuned for regular updates to these templates from our team.

To learn more about how NVIDIA is enhancing AI transparency and ethical considerations with model card, read our developer blog.