https://github.com/oniani/mnist
The Reproducible MNIST
https://github.com/oniani/mnist
ai mnist mnist-classification pytorch reproducibility
Last synced: 3 months ago
JSON representation
The Reproducible MNIST
- Host: GitHub
- URL: https://github.com/oniani/mnist
- Owner: oniani
- License: mit
- Created: 2022-10-19T01:40:49.000Z (almost 4 years ago)
- Default Branch: main
- Last Pushed: 2022-12-23T02:22:41.000Z (over 3 years ago)
- Last Synced: 2025-10-30T03:09:04.719Z (9 months ago)
- Topics: ai, mnist, mnist-classification, pytorch, reproducibility
- Language: Python
- Homepage:
- Size: 7.81 KB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# The Reproducible MNIST
Reproducible [MNIST](https://en.wikipedia.org/wiki/MNIST_database) experiments in
[PyTorch](https://pytorch.org/).
## Setup
```console
$ python -m venv .venv
$ source .venv/bin/activate
$ pip install --upgrade pip
$ pip install -r requirements.txt
$ wandb login API_KEY
```
## Model Training
```console
$ CUBLAS_WORKSPACE_CONFIG=:16:8 python train.py # or CUBLAS_WORKSPACE_CONFIG=:16:8 ./train.py
```
You should set the environment variable `CUBLAS_WORKSPACE_CONFIG` according to
[CUDA documentation](https://docs.nvidia.com/cuda/cublas/index.html#cublasApi_reproducibility).
Bonus: [WandB](https://wandb.ai/) takes care of generating the loss plot and saving the training
configuration.
## Model Evaluation
```console
$ python eval.py # or ./eval.py
```
## License
[MIT License](LICENSE)