tensorflow: model.evaluate() and tf.keras.losses.MSE returning completely different values

Question

I am training the following RNN for univariate time series forecasting.

After training, I want to evaluate the model with my test set. I tried to do that with both