Interactive reference for transformer models, presented via dataflow graphs, drillable down to elementary mathematical operations. Covers models from GPT-2 to Qwen 3.6, with MLA, MoE, RoPE, MTP, hybrid attention, and other variants toggleable in place.