modelcards

⚠️Deprecation Warning⚠️ Utils in this repo now live in huggingface_hub.

So, this project will no longer be maintained.

You can find more information in the Hugging Face Hub documentation:

• Walkthrough Guide
• Repo Cards API Reference Documentation

---

📝 Utility to create, edit, and publish model cards on the Hugging Face Hub.

For a full walkthrough, try the demo in Colab:

Usage

Installation

pip install modelcards

Examples

Load a model card from a Hugging Face Hub repo:

from modelcards import ModelCard

card = ModelCard.load("nateraw/food")

# Access its card data
print(card.data)

# Update its card data
card.data.library_name = "transformers"

# Save it to a file
card.save("my_card.md")

# Or, push it to the hub directly to replace the existing card
card.push_to_hub("nateraw/food")

Make model cards from the default model card template. 👀 You can see what the resulting model card looks like at this Hugging Face Hub repo.

from modelcards import CardData, ModelCard

repo_id = "nateraw/my-cool-model-with-card"

# Initialize card from default template, including card metadata
card = ModelCard.from_template(
    card_data=CardData(  # Card metadata object that will be converted to YAML block
        language='en',
        license='mit',
        library_name='timm',
        tags=['image-classification', 'resnet'],
        datasets='beans',
        metrics=['accuracy', 'f1'],
    ),
    model_id=repo_id.split('/')[-1],  # Jinja template kwarg
    model_description="Some really helpful description...",  # Jinja template kwarg
)

The modelcards.CardData class is used above to define some card metadata. This metadata is leveraged by the Hugging Face Hub to:

• enable discoverability of your model through filters
• provide a standardized way to share your evaluation results (which are then automatically posted to Papers With Code)
• enable the inference API if your model is compatible with one of the available Inference API pipelines.
• And more!

You can also make your own template and supply that to the from_template method by using the template_path argument.

from pathlib import Path

from modelcards import CardData, ModelCard

template_text = """
---
{{ card_data }}
---

# {{ model_id | default("CoolModel") }}

This model is part of `super_cool_models` package (which doesn't exist)! It is a fine tuned `cool-model` on the `{{ dataset_name }}`.

## Intended uses & limitations

This model doesn't exist, so you probably don't want to use it! This is just an example template. Please write a very thoughtful model card ❤️
"""

Path('my_template.md').write_text(template_text)

card = ModelCard.from_template(
    card_data=CardData(  # Card metadata object that will be converted to YAML block
        language='en',
        license='mit',
        library_name='super_cool_models',
        tags=['image-classification', 'cool-model'],
        datasets='awesome-dataset',
        metrics=['accuracy', 'f1'],
    ),
    template_path='my_template.md', # The template we just wrote!
    model_id='cool-model',  # Jinja template kwarg
    dataset_name='awesome-dataset', # Jinja template kwarg
)