A model card primarily documents what aspects?

A model card primarily documents what aspects?

• A. A deployment checklist.
• B. Model purpose, limitations, and performance.
• C. Training data sources only.
• D. Hardware requirements.

Answer: (B)

The question tests what information a model card is meant to convey about a model. The best description is that a model card documents the model’s purpose, its limitations, and its performance. This combination tells users why the model exists, what tasks it is intended to solve, what it can and cannot do, and how well it works on relevant evaluation metrics and benchmarks. By laying out purpose, constraints, and results, the card guides safe and appropriate use and helps stakeholders understand potential risks. Deployment checklists focus on operational steps to run and maintain the system, which is about deployment rather than the model’s behavior or capabilities. Training data sources alone provide information about data origins but don’t capture how the model behaves in practice, its strengths and weaknesses, or its performance. Hardware requirements center on the infrastructure needed to run the model, not on the model’s intended use or risk profile.

The question tests what information a model card is meant to convey about a model. The best description is that a model card documents the model’s purpose, its limitations, and its performance. This combination tells users why the model exists, what tasks it is intended to solve, what it can and cannot do, and how well it works on relevant evaluation metrics and benchmarks. By laying out purpose, constraints, and results, the card guides safe and appropriate use and helps stakeholders understand potential risks.

Deployment checklists focus on operational steps to run and maintain the system, which is about deployment rather than the model’s behavior or capabilities. Training data sources alone provide information about data origins but don’t capture how the model behaves in practice, its strengths and weaknesses, or its performance. Hardware requirements center on the infrastructure needed to run the model, not on the model’s intended use or risk profile.