LLM Architecture Gallery offers centralized technical documentation

An online gallery compiling architecture diagrams for language models, containing 93 model cards derived from 42 articles and technical reports. The gallery spans familiar dense architectures like GPT-2 and Llama 3, recent mixture-of-experts models like DeepSeek V3 and Kimi K2, and hybrids combining attention with recurrent or state-space layers. Each card presents a high-resolution diagram and fact sheet with technical details: model scale, context length, vocabulary size, license, decoder type, and attention mechanism. For models covered by Artificial Analysis, it includes Intelligence Index and performance profiles across general, scientific, coding, and agents categories.