Minimal PyTorch implementations of foundational deep learning papers. Each implementation focuses on clarity and educational value while remaining faithful to the original architecture.
| Paper | Directory | Description |
|---|---|---|
| Deep Residual Learning for Image Recognition | resnets/ |
ResNet-18 for image classification |
| Language Models are Unsupervised Multitask Learners | transformers/ |
GPT-2 decoder-only transformer |
- PyTorch >= 1.0
- NumPy
- Matplotlib
- tiktoken (for transformers)
Each implementation includes:
- Well-documented model code with paper references
- Training and evaluation utilities
- README with architecture overview and examples
See individual directories for detailed usage instructions.