Coupling Forget Gate and Input Gate of LSTM

Tensorflow provides a LSTM cell which couples the forget and input gate but otherwise acts a typical LSTM cell. Tensorflow also provides a GRU cell which is distinct.

I have been looking through the pytorch documentation and cannot seem to find any mention of a lstm cell of this type. Only GRU or LSTM’s but no variations on them.

I know the GRU combines these gates but I want the benefits of longer relation distances and to keep the cell state and hidden state separate.

Is there an equivalent in pytorch to the cell type provided by tensorflow or is this something I will need to build myself?

I would also be interested in this