← ~/visualizations

attention-mechanism #

Attention matrix showing how each token attends to other tokens. Query token row highlighted with attention weights.

canvasclick to interact

⏮◀◀▶▶STEP0.25x1xZOOM

t=0s

practical uses #

• 01.Teaching transformer architecture
• 02.Explaining self-attention
• 03.Visualizing token relationships
• 04.Demonstrating QKV computation

technical notes #

5x5 attention matrix with token labels. Active query row highlighted. Cell intensity = attention weight.

← dijkstra's-algorithmword-embeddings →