PyTorch-Implementations features implementations of foundational deep learning architectures. Ranging from classical computer vision and sequence-to-sequence NLP models to fundamental generative AI models, this repository serves as a reference for understanding the mechanisms of these architectures. All models are implemented natively in PyTorch and trained on standard benchmarking datasets.
PyTorch-Implementations/
├── DCGAN/ # Deep Convolutional Generative Adversarial Network
│ ├── models/ # Saved model (pth)
│ ├── results/ # Training loss plot
│ └── dcgan.ipynb # Implementation and training notebook
├── DDPM/ # Denoising Diffusion Probabilistic Model
│ ├── models/ │ ├── results/ │ └── ddpm.ipynb ├── Transformer/ # Sequence-to-Sequence Transformer
│ ├── models/ │ ├── results/ │ └── transformer.ipynb ├── UNet/ # UNet for Segmentation
│ ├── models/ │ ├── results/ │ └── unet.ipynb ├── Variational-Autoencoder/ # Variational Autoencoder
│ ├── models/ │ ├── results/ │ └── vae.ipynb ├── Vision-Transformer/ # Vision Transformer for Classification
│ ├── models/ │ ├── results/ │ └── vit.ipynb ├── .gitignore ├── LICENSE # MIT License
└── README.md # Project Documentation
| Domain | Specific Architecture | Dataset | Objective |
|---|---|---|---|
| Generative AI | Denoising Diffusion Probabilistic Model (DDPM) | CelebA | Unconditional Image Generation |
| Deep Convolutional GAN (DCGAN) | CelebA | Adversarial Image Generation | |
| Variational Autoencoder (VAE) | CelebA | Latent Space Reconstruction & Generation | |
| Computer Vision | Vision Transformer (ViT) | CIFAR-10 | Multi-class Image Classification |
| U-Net | ISBI 2016 | Biomedical Image Segmentation | |
| Natural Language | Transformer (Standard Seq2Seq) | Multi30k | English to German Translation |
| Architecture | Dataset | Task | Demonstration / Loss Curve |
|---|---|---|---|
| DDPM | CelebA | Image Generation |  |
| VAE | CelebA | Latent Generation |  |
| DCGAN | CelebA | Image Generation |  |
| U-Net | ISBI 2016 | Segmentation |  |
| Vision Transformer | CIFAR-10 | Classification |  |
| Transformer | Multi30k | Translation |  |
Follow these steps to set up the environment and run any of the models locally.
1. Clone the repository and install dependencies:
Ensure you have Python 3.10+ installed. Install the required libraries, including PyTorch and Torchvision:
git clone [https://github.com/amimayo/PyTorch-Implementations.git](https://github.com/amimayo/PyTorch-Implementations.git)
cd PyTorch-Implementations
pip install torch torchvision numpy matplotlib jupyter2. Navigate to the desired model:
Move into the directory of the specific architecture you want to explore.
cd DDPM3. Launch the Jupyter Notebook:
Open the notebook to view the architecture code, training loop, and inference results.
jupyter notebook ddpm.ipynb- [🟩] Core Architectures: Complete training and verification for UNet, Transformer, ViT, VAE, DCGAN, and DDPM.
- [🟨] BERT (Bidirectional Encoder Representations from Transformers): Implement masked language modeling for sequence representation.
- [🟨] GPT (Generative Pre-trained Transformer): Implement an autoregressive decoder-only architecture for text generation.
Distributed under the MIT License.