LLM Architecture & Token Transformer Simulator
D3.js Powered
Preset: Shakespeare
Preset: Python Code
Reset State
Export JSON
1. Input & Hyperparameters
Input Prompt Sequence:
To be or not to be
Attention Heads ($N_{heads}$):
4
Embedding Dim ($d_{model}$):
16
Sampling Temperature:
0.80
Top-K Sampling:
5
Tokenized Sequence
Status: Forward pass ready. Select or hover over tokens to inspect QK dot products & attention weight matrices.
2. Self-Attention Heatmaps ($Q \cdot K^T / \sqrt{d_k}$)
Attention Heatmaps
Vector Inspector
Token Embedding + Sinusoidal Positional Encoding breakdown for selected token:
3. Next-Token Logits & Sampling
⚡ Generate Next Token (Forward Pass)
Sequence Length
18
Softmax Entropy
1.42
Active Heads
4
Top Candidate
that
Sampled Probability Distribution:
Refocus Surface
🔍 Refocus Primary Proof Surface
Enjoy this tool? Build your own with Super