This message was deleted.

Your training procedure diverged, you may want to reduce the learning rate. You can monitor through this attribute models[0].train_trajectories https://github.com/Nixtla/neuralforecast/blob/main/neuralforecast/common/_base_windows.py#L85