FastGRNNCUDA Fix (#137)

* fixes for installation and fastgrnncuda

* ensure input tensors are on device

* ensure tensors on device for fastgrnncudacell

* add batch_first support

* fix forward params

* minor variable name fix

Author: MJ10

pytorch/edgeml_pytorch/graph/rnn.py

@@ -1170,7 +1170,7 @@ def forward(self, input, hiddenState, cell_state=None):
             input = input.to(self.device)
         if hiddenState is None:
             hiddenState = torch.zeros(
-                [input.shape[1], self.hidden_size]).to(self.device)
+                [input.shape[1], self._hidden_size]).to(self.device)
         if not hiddenState.is_cuda:
             hiddenState = hiddenState.to(self.device)
         return FastGRNNUnrollFunction.apply(input, self.bias_gate, self.bias_update, self.zeta, self.nu, hiddenState,