Custom LSTM cell implementation

I would like to implement a custom version of the typical LSTM cell as it is implemented in Pytorch, say, change one of the activation functions at a gate. For this, I would like to see how the LSTM is implemented in Pytorch at the moment. I can find some code here, but unfortunately, I cannot find the exact LSTM computations there etc.

Related posts can for example be found here, but all they delivered me is that nobody has found the LSTM cell code in github.

Where is the LSTM cell implemented in the Pytorch Github exactly?

class LSTMCell(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(LSTMCell, self).__init__()
        self.input_size = input_size
        self.hidden_size = hidden_size
        self.weight_ih = Parameter(torch.randn(4 * hidden_size, input_size))
        self.weight_hh = Parameter(torch.randn(4 * hidden_size, hidden_size))
        self.bias_ih = Parameter(torch.randn(4 * hidden_size))
        self.bias_hh = Parameter(torch.randn(4 * hidden_size))

    def forward(self, input, state):
        # type: (Tensor, Tuple[Tensor, Tensor]) -> Tuple[Tensor, Tuple[Tensor, Tensor]]
        hx, cx = state
        gates = (, self.weight_ih.t()) + self.bias_ih +
       , self.weight_hh.t()) + self.bias_hh)
        ingate, forgetgate, cellgate, outgate = gates.chunk(4, 1)

        ingate = torch.sigmoid(ingate)
        forgetgate = torch.sigmoid(forgetgate)
        cellgate = torch.tanh(cellgate)
        outgate = torch.sigmoid(outgate)

        cy = (forgetgate * cx) + (ingate * cellgate)
        hy = outgate * torch.tanh(cy)

        return hy, cy

This is what you r looking for.

1 Like