What AI says about AttentionVisualizer

AttentionVisualizer is a Python package that visualizes self-attention scores from encoder-only Transformer models (primarily RoBERTa) in IPython/Jupyter notebooks. It provides a lightweight UI to display the highest-scored words, with options to ignore tokens (BOS, punctuation, stopwords) and to inspect particular layers or attention heads, including a Google Colab demo. The project is hosted on GitHub, installable via pip from GitHub, and includes a demo notebook and a blog/post detailing the implementation.