Hi, you need to pass learn.crit = F.mse_loss for regression. By default RNN_Learner uses F.cross_entropy.