rak1213/Transformer-Interpretability

Transformer Interpretability using transformer lens. Plotting all the heads of a single layer to see different attentions and ablating different heads to see how much they affect output

Jupyter Notebook