Transformer Architecture Lab — Interactive Self-Attention & Deep Learning Explorer

Sentence Preset

Custom Input Text

Multi-Head Selector Head 1: Coreference

Softmax Temperature: 1.0

Sharper Focus (0.2) Diffused (3.0)

Inspect Query Token (Row)

Live Inference Engine Active

Projections: $d_{\text{model}} = 64$, $d_k = 16$. Math computed live in 64-bit vector space.