Comprehensive Gallery Catalogs 93 LLM Architectures

A large language model architecture gallery has expanded to document 93 different model designs across 42 technical sources, including dense architectures like GPT-2 and Llama 3, mixture-of-experts models such as DeepSeek V3 and Kimi K2, and hybrid approaches combining attention with recurrent or state-space layers. Each model card includes detailed architecture diagrams, technical specifications including context length and KV-cache estimates, and links to original papers and implementations.