Datasets:

---

orionweller

/

mmBERT-pretraining-data-chunk1

like 0

This PR enhances the dataset card for mmBERT Training Data by:

• Adding feature-extraction to task_categories: This reflects the primary utility of models trained on this data for downstream tasks like classification and retrieval, as described in the paper abstract and demonstrated in the GitHub repository's usage examples. The existing fill-mask category is retained as it represents a key pre-training objective.
• Adding a "Sample Usage" section: Incorporates relevant code snippets from the mmBERT GitHub repository, illustrating how models trained with this dataset can be used for tasks such as obtaining multilingual embeddings and performing multilingual retrieval. This helps users understand the practical applications of the data.
• Improving paper links: A new Hugging Face Paper badge is added, and the "Related Resources" section is updated to explicitly include a link to the Hugging Face paper page (https://huggingface.co/papers/2509.06888) alongside the existing Arxiv link, improving discoverability without replacing the original.
• Adding an introductory paragraph: Provides a concise overview of the dataset and its connection to the mmBERT paper.

These changes aim to make the dataset card more informative and user-friendly.