nn.Parameter和F.linear的用法

class TextRNN(nn.Module):
    def __init__(self, 
                 input_size = 768, 
                 hidden_size = 164,
                 output_size = 768,
                 n_layers = 1,
                 dropout =  0.0,
                 args = None
                 ):
        super(TextRNN, self).__init__()

        self.h0 = nn.Parameter(torch.Tensor(n_layers, hidden_size))
        self.c0 = nn.Parameter(torch.Tensor(n_layers, hidden_size))

        self.dropout = nn.Dropout(dropout) 

        self._activate = nn.Tanh()
    
        # block input
        self.Wz = nn.Parameter(torch.Tensor(input_size, hidden_size))  # [8, 768, 164]
        self.Rz = nn.Parameter(torch.Tensor(hidden_size, hidden_size)) # [8, 164, 164]
        self.bz = nn.Parameter(torch.Tensor(hidden_size, hidden_size)) # [8, 164, 164]

        # input gate
        self.Ai = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Wi = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Ri = nn.Parameter(torch.Tensor(hidden_size, hidden_size))
        self.Pi = nn.Parameter(torch.Tensor(n_layers, hidden_size))
        self.bi = nn.Parameter(torch.Tensor(n_layers, hidden_size))

        # forget gate
        # input_size = 768, hidden_size = 164
        self.Af = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Wf = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Rf = nn.Parameter(torch.Tensor(hidden_size, hidden_size))
        self.Pf = nn.Parameter(torch.Tensor(n_layers, hidden_size))
        self.bf = nn.Parameter(torch.Tensor(n_layers, hidden_size))

        # output gate
        self.Ao = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Wo = nn.Parameter(torch.Tensor(input_size, hidden_size))
        self.Ro = nn.Parameter(torch.Tensor(hidden_size, hidden_size))
        self.Po = nn.Parameter(torch.Tensor(n_layers, hidden_size))
        self.bo = nn.Parameter(torch.Tensor(n_layers, hidden_size))

    def forward(self, input_ids, input_attention):  # [8, 16