Skip to content

Dataset · MC7ever

Multimodal HF Corpus of 92K Datasets

MC7ever/hf-training-corpus

Large-scale extraction of texts, images, audio, and video from open datasets in a unified parquet format.

What's inside

A meta-dataset with a volume of 1 to 10 million English-language records, collected by automatically crawling about 92,000 datasets from Hugging Face. Different modalities are normalized and packed into parquet files: texts, images (PNG), audio (WAV), video fragments, and tabular data converted to text.

What you can do with it

  • Pretraining multimodal foundational models on heterogeneous data
  • Creating universal pipelines for training text generation, ASR, and image analysis models
Tasks
Text generation, Image → text, Automatic speech recognition
Languages
English
Size
100K–1M rows
Downloads this month
26.8K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("MC7ever/hf-training-corpus")