Hugging Face's logo

Datasets:
jhu-clsp
/
mmBERT-pretrain-p1-fineweb2-langs

Tasks:
Fill-Mask
Feature Extraction
Languages:
multilingual
ArXiv:
Tags:
pretraining
language-modeling
encoder
multilingual
feature-extraction
License:
Dataset card Files
xet
Community
1

Update dataset card metadata (language, task categories, tags) and add HF paper link

#1
by nielsr HF Staff - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+4
-1
Update dataset card metadata (language, task categories, tags) and add HF paper linkc5ad3f32
nielsr

This PR updates the dataset card for mmbert-pretrain-p1-fineweb2-langs to better reflect its characteristics and utility.

Specifically, it:

  • Updates the language metadata from en to multilingual to accurately represent the dataset's coverage of 60 languages in this phase and its role in training models for over 1800 languages.
  • Adds feature-extraction to the task_categories and tags. While fill-mask represents a pre-training objective, the mmBERT models trained on this data are primarily used for general-purpose feature extraction, which underpins tasks like classification and retrieval.
  • Adds a direct link to the Hugging Face paper page (https://huggingface.co/papers/2509.06888) under the "Related Resources" section for improved discoverability, while retaining the existing Arxiv badge.
orionweller changed pull request status to merged

Sign up or log in to comment