Calculating accuracy using model.evaluate yields different value then that calculated manually giving prediction and ground truth in tensorflow keras

Question

Please note that I am trying to apply some time series forecasting in this project.

I have trained the following model:

forecast_period = 6
tf.keras.ba