Open nankepan opened 2 months ago
One more question: Is the cosine similarity calculation based on flattened latent or other ways? Could the author share the calculation code? I would like to verify the redundancies mentioned in the paper. Thanks!
We have added the code to calculate and visualize the similarity of attention outputs in experiments/cfg_similarity_experiments/. Please comment below if you encounter any issues while running the code or if you have further questions about the output.
Hi, Very nice work! How to generate visual image of 'Similarity of Attention Outputs' as shown in Figure 4? Could the author share the code for visualization??