Mixture-of-experts

Published on
March 6, 2025
DeepSeek-R1-Zero and DeepSeek-R1: Reinforcement Learning & Fine-Tuning Analysis
A comprehensive breakdown of DeepSeek-R1-Zero and DeepSeek-R1, covering Reinforcement Learning (RL), Supervised Fine-Tuning (SFT), architecture, and performance improvements.
Published on
March 3, 2025
Demystifying DeepSeek-V3: Breaking Down Its Revolutionary AI Architecture
Breaking down DeepSeek-V3’s revolutionary AI architecture—exploring its key innovations, expert routing, and inference optimizations step by step. This post dives deep into the mathematics and mechanisms that power its efficiency and scalability.