This PR enhances the dataset card for the DCLM pre-training data used in MoCa models.
It adds the `task_categories: feature-extraction` to the metadata, aligning with the dataset's purpose in creating bidirectional multimodal embeddings.
The `tags` are refined to include `multimodal`, `embeddings`, and `vision-language`, which better describe the dataset's role and improve its discoverability on the Hugging Face Hub.
Additionally, a "Sample Usage" section has been added to guide users on how to use the models trained with this dataset for inference, referencing the associated code repository and its `demo.py` script.
