self_attention python代码

from math import sqrt
import torch
import torch.nn as nn

class SA(nn.Module):
    def __init__(self, dimQ, dimK, dimV):
        super(SA, self).__init__()

        self.dimQ = dimQ
        self.dimK = dimK
        self.dimV = dimV

        # self.mid = 10
        
        self.linerQ = nn.Linear(self.dimQ, self.dimV, bias = False)
        self.linerK = nn.Linear(self.dimK, self.dimV, bias = False)
        self.linerV = nn.Linear(self.dimV, self.dimV, bias = False)

        self.sqrtD = 1 / sqrt(dimQ)
    
    def forward(self, x):
        batch, n, dim = x.shape

        assert(dim == self.dimQ)

        Q = self.linerQ(x)
        K = self.linerK(x)
        V = self.linerV(x)

        dist = torch.bmm(Q, K.transpose(1, 2)) * self.sqrtD
        W = torch.softmax(dist, dim = -1)

        Output = torch.bmm(W, V)
        return Output

if __name__ == "__main__":
    x = torch.tensor([[[1,2,3],[2,3,4],[3,4,5],[4,5,6]],
                     [[1,2,3],[2,3,4],[3,4,5],[4,5,6]]], dtype = torch.float)
    print(x.shape)

    saModel = SA(3, 3, 3)
    Output = saModel(x)
    print(Output)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

https://zhuanlan.zhihu.com/p/338817680
https://blog.csdn.net/weixin_44750512/article/details/124244915
https://blog.csdn.net/qq_40178291/article/details/100302375