Forward() takes 2 arguments but 3 were given

autograd
3

nn.Sequential container use a single input and output tensor as the input and output activation.
You could write a custom nn.Module for multiple inputs or check e.g. this topic for more information and potential workarounds.