Open training data
QVAC Genesis
The data to train your own models. An open dataset of 339 billion tokens of synthetic STEM reasoning, built so independent researchers can train reasoning-capable models of their own.
from datasets import load_dataset
ds = load_dataset("qvac/GenesisIII", split="train")
print(ds[0])
# level-aligned Q&A with reasoning traces
In the open
The fuel big labs hoard, in the open
Reasoning-capable models need reasoning-rich data. Genesis puts it in the open so independent researchers can train models of their own.
Built for reasoning
Level-aligned questions with reasoning traces across 19 STEM domains.
Open for research
CC-BY-NC 4.0, free for research and education.
One line to load
Hosted on HuggingFace, loaded with the datasets library.
Three releases
339 billion tokens, 19 STEM domains
Three open releases on HuggingFace, all CC-BY-NC 4.0 for research and education.
0B
tokens of open synthetic STEM reasoning data
Quickstart
Load it in one line
Hosted on HuggingFace. Stream it with the datasets library, or pull it locally for training runs.
# pull Genesis III locally
huggingface-cli download \
qvac/GenesisIII \
--repo-type dataset \
--local-dir ./genesis-iii
# Genesis I: 41B tokens, 4 domains
# Genesis II: 107B tokens, 19 domains
# Genesis III: 191.43B tokensGet started
Run AI on what you own
One install, JavaScript or Python. 10+ AI tasks. Private, offline and free on the hardware you already own.