What is model documentation (model card, datasheet)?

Model documentation, including model cards and datasheets, are standardized documents that describe the characteristics, capabilities, limitations, and usage conditions of machine learning models.

• A model card typically contains information about the model's purpose, used data, quality metrics, potential risks, fairness and ethical considerations. It helps users and developers understand how and where the model can be applied, and identify possible sources of bias.

• A datasheet for a model is a more technical document that details the architecture, training process, datasets used for training and testing, and experimental results.

Such documentation increases transparency, fosters trust in the model, and helps assess its suitability for specific tasks, especially in critical areas where fairness and safety are paramount.