CodeThatPaper: Research Paper Reproductions & Architecture Implementations

This repository contains bare-metal implementations and reproductions of foundational Artificial Intelligence research papers. The goal of this project is to deconstruct state-of-the-art architectures from theoretical equations into highly optimized, modular code across PyTorch, TensorFlow, and NumPy.

Detailed reproduction metrics, original paper comparisons, and execution scripts are located in the respective sub-directories.

📌 Architecture Directory

Vision & Multi-Modal Models

These implementations focus on attention mechanisms and feature extraction in computer vision, aligning with cross-domain adaptation tasks.

• Vision Transformer (ViT) - An Image is Worth 16x16 Words
• AlexNet - Deep Convolutional Neural Networks (TensorFlow)

Large Language Models & Attention Mechanisms

From the foundational Transformer to modern causal language models.

• Attention is All You Need - Core Transformer architecture from scratch (PyTorch)
• LLaMA (Small Edition) - Reproduction of the LLaMA architecture (PyTorch)
• GPT - Generative Pre-trained Transformer from scratch (PyTorch)
• LoRA/ QLoRA - LoRA and QLoRA from Scratch for Vision Language Model Fine Tuning (PyTorch)

Generative Adversarial Networks

• Conditional DCGAN (CGAN) - (PyTorch)
• Deep Convolutional GAN (DCGAN) - (PyTorch)
• Standard GAN - (TensorFlow)

Core Operations & Fundamentals

• TurboQuant - Low-level quantization implementation (NumPy)
• Fully Connected Network - Dense layer math from scratch (NumPy)

---

Connect for research discussions or view my enterprise AI experience on LinkedIn.