DeepSeek Exploration

Welcome to DeepSeek Exploration — a comprehensive deep dive into DeepSeek, where we investigate its architecture, inner workings, and key components inside out. This project aims to help researchers, developers, and enthusiasts understand how DeepSeek operates, enabling better usage, modification, and innovation.

Table of Contents

1. Self-Attention Mechanism
2. Causal-Attention Mechanism
3. Multi-Head Attention Mechanism
4. Normalization Techniques
5. KV Cache Mechanism
6. Position Embbeddings

more ......