---
license: unknown
pretty_name: CIFAR-10 (original python archive)
task_categories:
- image-classification
---
# CIFAR-10, original python archive

Unmodified copy of `cifar-10-python.tar.gz` from https://www.cs.toronto.edu/~kriz/cifar.html,
mirrored so course notebooks do not depend on the original host.

- MD5 `c58f30108f718f92721af3b95e74349a` (the value `torchvision.datasets.CIFAR10` checks)
- 170,498,071 bytes

## Use with torchvision

Set the URL before the first `datasets.CIFAR10(...)` call. torchvision still verifies the
MD5, so nothing else changes. Pin a commit hash from this repo's history, not `main`.

    datasets.CIFAR10.url = ("https://huggingface.co/datasets/MIT-OL-AI-D/cifar-10-python/"
                            "resolve/<commit>/cifar-10-python.tar.gz")

## Citation

Alex Krizhevsky, *Learning Multiple Layers of Features from Tiny Images*, 2009.