5fa1a76
1
2
3
Next, the feature map is flattened and transposed to obtain a tensor of shape (batch_size, seq_len, d_model) = (batch_size, width/32*height/32, 256).
