Model Card

Model Summary

Provide a brief overview of the model including details about its architecture, how it can be used, characteristics of the model, training data, and evaluation results.

Usage

How can this model be used? You should provide a code snippet that demonstrates how to load and/or fine-tune your model, and you should define the shape of both the inputs and the outputs. Are there known and preventable failures to be aware of?

System

Is this a standalone model or part of a system? What are the input requirements? What are the downstream dependencies when using the model outputs?

Implementation requirements

What hardware and software were used for training the model? Describe the compute requirements for training and inference (e.g., # of chips, training time, total computation, measured performance, energy consumption).

Model Characteristics

Model initialization

Was the model trained from scratch or fine-tuned from a pre-trained model?

Model stats

What’s the size of the model? Provide information about size, weights, layers, and latency.

Other details

Is the model pruned? Is it quantized? Describe any techniques to preserve differential privacy.

Data Overview

Provide more details about the data used to train this model.

Training data

Describe the data that was used to train the model. How was it collected? What pre-processing was done?

Demographic groups

Describe any demographic data or attributes that suggest demographic groups

Evaluation data

What was the train / test / dev split? Are there notable differences between training and test data?

Evaluation Results

Summary

Summarize and link to evaluation results for this analysis.

Subgroup evaluation results

Did you do any subgroup analysis? Describe the results and any assumptions about disaggregating data. Are there any known and preventable failures about this model?

Fairness

How did you define fairness? What metrics and baselines did you use? What were the results of your analysis?

Usage limitations

Are there sensitive use cases? What factors might limit model performance and what conditions should be satisfied to use this model?

Ethics

What ethical factors did the model developers consider? Were any risks identified? What mitigations or remediates were undertaken?