metadata
language: en
license: cc0-1.0
tags:
- mnist
- tiny-model
- early-stopping
Tiny MNIST Classifier
- Parameters: 970 (<1000)
- Test accuracy: 92.35%
- Epochs trained: 45 (early stopping after 5 epochs without improvement)
This model was trained on RX 6600.
Full results
| Metric | Value |
|---|---|
| Total parameters | 970 |
| Best validation loss | 0.2463 |
| Final test accuracy | 92.35% |
| Early stopping patience | 5 |
| Training epochs | 45 |
Model architecture
AvgPool(4x4) → Linear(49→16) → ReLU → Dropout(0.2) → Linear(16→10)
How to use
import torch
from train import TinyMNISTModel
model = TinyMNISTModel()
model.load_state_dict(torch.load("mnist_1k_best.pth"))
model.eval()