After reviewing your paper, I identified the separate attention structure as one of the key design elements. As I understand it, this structure appears to be implemented by appropriately adjusting the attention map based on geometric criteria.
However, when examining the currently released code and debugging it, I observed that the attn map variable is always being passed as None. I would like to ask whether this behavior is intentional.
If this is not intentional, I would also appreciate clarification on how the separate attention mechanism is expected to be implemented in practice.
Thank you in advance for your help.
After reviewing your paper, I identified the separate attention structure as one of the key design elements. As I understand it, this structure appears to be implemented by appropriately adjusting the attention map based on geometric criteria.
However, when examining the currently released code and debugging it, I observed that the attn map variable is always being passed as None. I would like to ask whether this behavior is intentional.
If this is not intentional, I would also appreciate clarification on how the separate attention mechanism is expected to be implemented in practice.
Thank you in advance for your help.