Welcome to my repository, where I document my journey through the core concepts of machine learning by implementing models and tools from scratch. This serves as both a personal reference and an educational resource for anyone interested in understanding ML from first principles.
- Description: A basic sequential neural network capable of multiclass classification.
- Dataset: Tested on the MNIST dataset using a 3-layer approach, with ReLU as the activation function for hidden layers.
- Features: Implements forward propagation, backpropagation, and gradient descent from scratch using only Python and NumPy.
- Description: A Transformer model designed for English-to-Spanish translation.
- Tokenizer: Utilizes BPE Tokenization with a vocabulary size of 10000.
- Features: Implements attention mechanisms and positional encodings to achieve high-quality translations.
- Sequential Neural Network: Implemented and tested.
- Transformer-Based NLP Model: Successfully implemented and deployed.
- Retrieval-Augmented Generation (RAG): Integrating retrieval-based techniques with generative models.
- Reinforcement Learning from Human Feedback (RLHF): Exploring fine-tuning of LLMs with human preferences.
- Python: Core programming language for all implementations.
- NumPy & Pandas: Essential for data processing and matrix operations.
- PyTorch: For deep learning model development.
- Rust (Planned): Exploring performance improvements for certain implementations.
I’ll continue adding new implementations and optimizations as I learn more. If you're interested in ML from scratch, feel free to explore, contribute, or follow along!