replace forget_gate * cell_gate with input_gate * cell_gate

[seh2bp]

def lstm_cell(
    input: Tensor,
    hidden: tuple[Tensor, Tensor],
    w_ih: Tensor,
    w_hh: Tensor,
    b_ih: Tensor,
    b_hh: Tensor,
) -> tuple[Tensor, Tensor]:
    hx, cx = hidden
    gates = torch.mm(input, w_ih.t()) + torch.mm(hx, w_hh.t()) + b_ih + b_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) + (forgetgate * cellgate)
+   cy = (forgetgate * cx) + (ingate * cellgate)
    hy = outgate * torch.tanh(cy)

    return hy, cy

[ArmRyan]

Hi @seh2bp , Thank you for the contribution!

I am not very familiar with LSTM so I will just need to verify the change and run the patch through our CI also

[ArmRyan]

Hi @seh2bp I ran this through our ci and its all good, the only thing missing is the versions on both files need to be updated and also the dates, then I am good to approve it!

I am wondering though if you had a problem with the function before? The forget gate and input gate both point to the same buffer so the output should be the same. I think this is still more clear so I think it is a good change, but if you had a problem is shouldnt fix it?

[ArmRyan]

• $Date: 26 March 2024
• $Revision: V.1.0.0

Should be updated to current date and version should be V.1.0.1 in both files

[seh2bp]

Hi @seh2bp I ran this through our ci and its all good, the only thing missing is the versions on both files need to be updated and also the dates, then I am good to approve it!

I am wondering though if you had a problem with the function before? The forget gate and input gate both point to the same buffer so the output should be the same. I think this is still more clear so I think it is a good change, but if you had a problem is shouldnt fix it?

Updated the revision number & date.
I changed the code for my test before checking whether the original works.
I agree it was not a bug, but it did confuse me, and also agree that with the change further confusion might be prevented.

[ArmRyan]

I changed the code for my test before checking whether the original works. I agree it was not a bug, but it did confuse me, and also agree that with the change further confusion might be prevented.

Ok great, just wanted to make sure that there was no other bug to deal with! Thanks for the contribution :)