Data as of Jul 25, 2026 · Based on 2,718,867 AI responses across 9,511 prompts · See how Parse measures this
BERT Attention Visualizationgithub.com · Low evidenceBertAttentionViz is a Streamlit-powered web application that visualizes attention weights from BERT-based models (DistillBERT). Users input text, choose layers and heads, and view attention visualizations, with options to exclude special tokens and plot top predictions for masked tokens in multiple formats. The project includes a live Streamlit demo and can be run locally by cloning the repository and running the Streamlit app (streamlit_app.py).
Parse Score