Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attention

Open in new window