README.md

# AutoEncoder

A simple autoencoder trained on MNIST.

This model is part of the "Introduction to Generative AI" course.  
For more details, visit the [GitHub repository](https://github.com/hussamalafandi/Generative_AI).

## Model Description

The AutoEncoder is a neural network designed to compress and reconstruct input data. It consists of an encoder that compresses the input into a latent space and a decoder that reconstructs the input from the latent representation.

## Training Details

- **Dataset**: MNIST (handwritten digits)
- **Loss Function**: Mean Squared Error (MSE)
- **Optimizer**: Adam
- **Learning Rate**: 0.001
- **Epochs**: 40
- **Latent dim**: 10

## Tracking

For detailed training logs and metrics, visit the [Weights & Biases run](https://wandb.ai/hussam-alafandi/mnist-autoencoder/runs/f81c7dgf?nw=nwuserhussamalafandi).

## Load Model

```python
from model import AutoEncoder
import torch

model = AutoEncoder()
model.load_state_dict(torch.load("model.pth"))
model.eval()
```


## License

This project is licensed under the MIT License. See the LICENSE file for details.