Dataset · HuggingFaceFW
Educational PDF Tokens in 69 Languages
HuggingFaceFW/finepdfs-edu
350B+ educational PDF tokens filtered by an educational quality classifier.
What's inside
The corpus contains over 350 billion tokens from PDF documents selected for educational quality. Filtering was performed by a classifier trained on Qwen3-235B-A22B-Instruct-2507 annotations for 69 languages, using FinePDFs as the source dataset. Data is provided in tabular and text formats.
What you can do with it
- Pretrain LLMs on educational texts
- Fine-tune models for STEM and academic tasks
- Evaluate the quality of educational corpora and filters
- Tasks
- Text generation
- Languages
- English, German, Japanese, French, Spanish, Italian, Russian, Portuguese and 4 more
- Size
- 10M–100M rows
- Downloads this month
- 20.9K
- Likes
- 107
Models trained on it
- ceia-research/lumi-4bText generation 4.9K
- HPLT/NorOLMo-13BText generation 1.2K
- nkthebass/tinybrainbot-53m-a9m-hybrid-instructText generation 1K
- ufakai/ufakzeka-1-baseText generation 863
- mradermacher/Llama-Poro-2-8B-Long-Base-i1-GGUF 834
- HPLT/FinOLMo-13BText generation 811
- nkthebass/tinybrainbot-25m-v5.1-baseText generation 745
- nkthebass/tinybrainbot-360m-a75m-v4.1-moe-baseText generation 587
How to load it
from datasets import load_dataset
ds = load_dataset("HuggingFaceFW/finepdfs-edu")