jhu-clsp

/

mmBERT-checkpoints

like 3

Follow

Center for Language and Speech Processing @ JHU 190

This PR significantly enhances the model card for the mmBERT raw training checkpoints by:

• Updating the language tag to mul to accurately reflect its multilingual nature (over 1800 languages).
• Adding pipeline_tag: feature-extraction to improve discoverability on the Hub and enable the automated widget. While this repository contains raw checkpoints, the associated models are commonly used for feature extraction.
• Specifying library_name: transformers as the linked GitHub repository provides extensive usage examples with the transformers library.
• Adding descriptive tags like bert, multilingual, and encoder-only for better searchability.
• Updating the main title to reflect the paper's title and linking to the Hugging Face paper page.
• Incorporating the paper's abstract.
• Including comprehensive usage examples from the GitHub README's "Quick Start" and "Getting Started" sections, showcasing how to use the associated mmBERT models (e.g., mmbert-small, mmbert-base) for various tasks like feature extraction, masked language modeling, classification, and retrieval.
• Clarifying the purpose of this repository as containing raw training checkpoints and directing users to the model collection for runnable models.
• Integrating other valuable sections from the GitHub README such as "Model Family", "Training Details", "Evaluation", "FAQ", and "Limitations" to provide a complete overview of the project.

These changes make the model card more informative, discoverable, and user-friendly on the Hugging Face Hub.