Datasets:

---

orionweller

/

mmBERT-pretrain-p2-fineweb2-remaining

like 0

This PR updates the dataset card to be more comprehensive and accurate:

• Metadata Updates:
  • Changed language from en to mul to correctly indicate the dataset's multilingual nature, supporting over 1800 languages for model pre-training.
  • Expanded task_categories to include feature-extraction, text-classification, and text-retrieval, in addition to fill-mask, reflecting the primary downstream applications for which models trained on this data are used, as described in the paper abstract and GitHub README.
• Content Enhancements:
  • Added the explicit paper title for better discoverability.
  • Integrated detailed "Sample Usage with mmBERT Models" from the GitHub README, demonstrating how to use models trained on this data for feature extraction, masked language modeling, classification, multilingual retrieval, and long context processing. This makes the dataset card much more practical for users.
  • Renamed the existing "Usage" section to "Data Usage" to differentiate it from the newly added model usage examples.

These changes provide a clearer and more informative overview of the dataset and its utility for the community.