Closed Innary closed 1 year ago
thank you for your models' code! attention.py#LL381C2-L381C2 Why not use the normal softmax(attn_output_weights, dim=-1) in attention.py instead of subtracting the max value?
normal softmax(attn_output_weights, dim=-1)
https://github.com/IDEA-Research/DAB-DETR/issues/48
Innary
commented
1 year ago
thank you for your models' code! attention.py#LL381C2-L381C2 Why not use the
normal softmax(attn_output_weights, dim=-1)in attention.py instead of subtracting the max value?