gru

Are torch.nn.ReLU and torch.nn.Sigmoid trainable? Question: I build a simple GRU model with PyTorch. It includes 4 sub-modules. I noted that some dictionaries return by the state_dict() of them are empty after training, while ones of the other sub-modules certainly have some weights and bias. The code: class GruModel(nn.Module): def __init__(self, inputs, nodes, layers=2): super(GruModel, …