Vision Transformer (ViT) from Scratch

This repository contains a PyTorch implementation of the Vision Transformer (ViT), built completely from scratch without relying on high-level Transformer libraries.
It follows the approach introduced in the paper:

📄 An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Alexey Dosovitskiy et al., ICLR 2021

---

🚀 Features

• Custom implementation of core ViT components:
  • Patch Embedding
  • Multi-Head Self Attention (MHSA)
  • Position Embeddings
  • Transformer Encoder Layers
  • Classification Head
• Training pipeline for CIFAR-10 dataset

---

📂 Project Structure

vision_transformers_from_scratch/
│── dataset.py 
│── train.py 
│── models/
│ └── ViT.py 
│── best_vit_cifar10.pth