A model card is documentation that explains how a machine-learning model was built, tested, and evaluated. It helps users and stakeholders understand the model’s intended uses, limitations, and risks. section 3.18
Model cards support responsible governance by making development and evaluation more visible. They are especially useful where training data may contain bias, model outputs may expose private details, or organizations need to assess risks across technical, legal, policy, and ethical perspectives. section 3.18
A model card can help people assess:
- Whether a model fits its intended use. section 3.18
- What limitations, biases, and risks may affect its outputs. section 3.18
- How the model’s development and evaluation were conducted. section 3.18
Llama model cards
Meta released Llama models with detailed model cards describing how they were built. Their release contributed to debate about open access, safeguards, and responsible artificial-intelligence release practices. section 1.18
