gpu_nums> 1

If want to run on multi gpu, when self.shared is forwarding , should use Modulelist's data (like self._w_h(which is a type of ModuleList)).
Otherwise will raise an error :( RuntimeError: tensors are on different GPUs) , beacuse when self.forward(xx), the parameter are used stored in list data structure, and would not replicate to another gpu.

Locate the self.shared forwarding path and inspect how self._w_h and other ModuleList data are accessed. Reproduce the issue with gpu_nums > 1 and a forward pass, then verify that multi-GPU execution no longer raises a tensors-on-different-GPUs error.

Written by the indexing model from the issue text.