Transformer Studio
Visual scaled dot-product attention, multi-head projection & next-token sampling
Live Transformer Lab
Preset Prompt
Fox & Dog (Full Syntax)
Attention is all you need
Pronoun Disambiguation ("it")
Parallel Computation
Custom Text
Softmax Temp (
1.0
)
Attention Head
H1: Semantic
H2: Syntactic
H3: Positional
H4: Recency
1. Multi-Head Self-Attention Heatmap (Q × Kᵀ / √dₖ)
Click cell to inspect math
2. Live Attention Inspector
Select any token or matrix cell to trace calculations.
Query (Q):
-
[--]
Key (K):
-
[--]
3. Next-Token Output Distribution
⚡ Generate Token