SubLayers.py 文件源码

def __init__(self, n_head, d_model, d_k, d_v, dropout=0.1):
        super(MultiHeadAttention, self).__init__()

        self.n_head = n_head
        self.d_k = d_k
        self.d_v = d_v

        self.w_qs = nn.Parameter(torch.FloatTensor(n_head, d_model, d_k))
        self.w_ks = nn.Parameter(torch.FloatTensor(n_head, d_model, d_k))
        self.w_vs = nn.Parameter(torch.FloatTensor(n_head, d_model, d_v))

        self.attention = ScaledDotProductAttention(d_model)
        self.layer_norm = LayerNormalization(d_model)
        self.proj = Linear(n_head*d_v, d_model)

        self.dropout = nn.Dropout(dropout)

        init.xavier_normal(self.w_qs)
        init.xavier_normal(self.w_ks)
        init.xavier_normal(self.w_vs)