🤖 Word-Level Attention Visualizer — choose a model & explore

Generate text, then select a generated word to see where it attends. The viewer below shows attention over both the prompt and the generated continuation. EOS tokens are stripped so <|endoftext|> doesn’t appear.

0) Model
Causal LM

Models that work with AutoModelForCausalLM + attentions

1) Generation
10 200
0 1.5
0 1

2) Attention
0 11
0 11
Select Generated Word

Selector lists only generated words
Attention visualization will appear here.