attention代码

s计算score的方式有以下几种：

import torch
class Attn(nn.module):
    def __init__(self, method ,hidden_size):
        super(Attn, self).__init__()
        self.method = method
        self.hidden_size = hidden_size
        if self.method not in ['dot', 'general', 'concat']:
            raise ValueError(self.method, "is not an appropriate attention method.")
        self.hidden_size = hidden_size
        if self.method == 'general':
            self.attn = nn.Linear(self.hidden_size, hidden_size)
        elif self.method == 'concat':
            self.attn = nn.Linear(self.hidden_size * 2, hidden_size)
            self.v = nn.Parameter(torch.FloatTensor(