A tool to visualize attention weights on a document, adapted from the attention visualizer by Abigail See. It looks like this when deployed:
Run this in the root directory:
python -m http.server
And go to http://localhost:8000/
in your browser. Text and weights will be loaded from a attentions.json
file, which contains a list of examples, with each example containing 4 fields:
words
: a list of words in the text being classifiedweights
: a list of weights assigned to each wordlabel
: the gold label of the textprediction
: the predicted label