| license: mit | |
| tags: | |
| - pytorch | |
| - mnist | |
| - modal | |
| # MNIST Model Trained on Modal | |
| This is a PyTorch model trained on the MNIST dataset using an NVIDIA A100 GPU on [Modal](https://modal.com). | |
| ## Model Details | |
| - **Architecture**: Simple CNN (Conv2D -> Conv2D -> MaxPool -> Dropout -> FC -> Dropout -> FC) | |
| - **Dataset**: MNIST | |
| - **Final Test Accuracy**: 99.02% | |
| ## Links | |
| - **W&B Run**: [View training on Weights & Biases](https://wandb.ai/ivanleo97-freelance/mnist-modal/runs/tu4yqtvi) | |
| - **Training Script**: The script used to train this model is available in the repository as `train_mnist.py`. | |
| ## Training Configuration | |
| - **Batch Size**: 64 | |
| - **Learning Rate**: 1.0 (Adadelta) | |
| - **Epochs**: 5 | |
| ## Training Metrics | |
| | Epoch | Train Loss | Test Loss | Accuracy | | |
| |:---:|:---:|:---:|:---:| | |
| | 1 | 0.2036 | 0.0441 | 98.40% | | |
| | 2 | 0.0828 | 0.0376 | 98.84% | | |
| | 3 | 0.0622 | 0.0487 | 98.38% | | |
| | 4 | 0.0528 | 0.0301 | 99.05% | | |
| | 5 | 0.0466 | 0.0314 | 99.02% | | |