Presets:
d_model (Embed Dim)
64
Heads Count (h)
4
Temperature (τ)
1.0
Focus Query Token
Scaled Dot-Product Attention Heatmap
[Q · K^T / √(d_k)]
Token Context Flow Graph
Routing per Head
Mathematical Vector Projection & Attention Breakdown
1. Selected Query Token Projection (Q)
Select or hover a token
2. Raw Dot-Products (Q · K^T / √(d_k))
Calculated scores...
3. Softmax Distribution α
Softmax probabilities...
4. Synthesized Context Vector (Σ α_i V_i)
Aggregated representation...
✓ Execution State Validated
Tokens: 0
d_k: 16
Selected Head: 0
Focused Query: none
Top Attention Target: -