Dataset · PleIAs
Synthetic corpus for reasoning models
PleIAs/SYNTH
79.6 million Wikipedia-based text examples for end-to-end training of small reasoning models.
What's inside
A synthetic text dataset for end-to-end training of small reasoning models. Contains 79,648,272 text samples, over 41 billion words (about 75 billion tokens). Collected by Pleias and AI Alliance based on 58,698 articles from Structured Wikipedia; languages are en, fr, it, es, de, pl.
What you can do with it
- End-to-end training of a small reasoning model
- SFT and synthetic data experiments
- Testing summarization and zero-shot classification
- Tasks
- Text generation, Zero shot classification, Summarization
- Languages
- English, French, Italian, Spanish, German, Polish, Dutch, Latin
- Size
- 10M–100M rows
- Downloads this month
- 37.2K
- Likes
- 312
Models trained on it
- PleIAs/MonadText generation 4.9K
- PleIAs/BaguettotronText generation 2.4K
- PleIAs/baguettotron-600mText generation 848
- PleIAs/Baguettotron-MoEText generation 279
- mkurman/NeuroBLAST-V3-SYNTH-EC-150000Text generation 67
- Ma7ee7/gemma-3-1b-thinking-gguf-F16Text generation 59
- mkurman/NeuroBLAST-V3-0.6B-SYNTH-EC-144B-TOKText generation 56
- mkurman/ConvGPT-0.2B-SYNTH-250B-ECText generation 50
How to load it
from datasets import load_dataset
ds = load_dataset("PleIAs/SYNTH")