LLM Visualizations

Ten interactive views of a language model end to end — from raw text to a sampled token. Each runs in the browser; nothing to install.

01. High-Level LLM Flow

02. Tokenization

03. Word Embeddings

04. Self-Attention Matrix

05. Multi-Head Attention

06. QKV Interaction

07. Feed Forward Network

08. Logits Waterfall

09. Softmax Distribution

10. Decoding Sampler