Token Travel
how GPT-2 small moves each token through its 13 layers
Run
Preset
Backend
ONNX (browser)
Server (FastAPI)
Projection
Raw
Per-layer standardized
Token
all
Layer
all
Animate
Reset view
Show attention
Head
avg
0
1
2
3
4
5
6
7
8
9
10
11
Threshold
0.05
Patch token
— none —
Replace with
Run patched
Loading…
heatmap sentence
—