Visualizing query-key interactions in language + vision transformers
- Preprint: https://arxiv.org/abs/2305.03210
- Demo: http://attentionviz.com/
- Docs: https://catherinesyeh.github.io/attn-docs/
- Clone repo and navigate into folder:
git clone https://github.com/catherinesyeh/attention-viz.git
cd attention-viz
- Ask Catherine for
data
folder. It should be included in theweb
folder like so:
- Navigate to back end:
cd web/back/
- Create virtual env and activate:
python3 -m venv env
source env/bin/activate
- Install requirements:
pip3 install -r requirements.txt
- Start back end:
python3 run.py
- Navigate to front end:
cd ../front
- Install necessary packages and start front end:
npm i
npm run serve
- The interface should be running at: http://localhost:8561