LLM Architecture Gallery

This page catalogs architecture diagrams and fact sheets for major open-weight LLMs, covering models from Llama 3 8B through Qwen3 Next 80B. Key architectural trends include widespread adoption of Sparse MoE designs (DeepSeek V3/R1, Qwen3 235B, GLM-4.5, GPT-OSS), GQA-based attention variants with techniques like RoPE, QK-Norm, and sliding-window/global alternation, and a general push toward serving efficiency at scale.