How can I use dot-product attention on two cnn features?

vision
1

there are two features from two CNN, and they share the same parameters, in my case, its shape of <128, 764>. I want to add dot-product attention on them, how can I implement them in PyTorch.

below picture contains my attention score function:

image

and W denotes the learning parameters.

2 
    def attention_cnn(self, c, q):
        ......
        alpha = F.softmax(u, -1)  # batch_size, concept_seq_len
        
        return alpha

framework maybe like this, alpha is the attention weight.