[bug] Model training much more slowly after Lightning Migration

[Kevin-Chen0]

Using the autoregression_yosemite_temps.ipynb and the following code as the example:

%%time
m = NeuralProphet(
    n_lags=12,
    changepoints_range=0.95,
    n_changepoints=30,
    weekly_seasonality=False,
)
metrics = m.fit(df)

in v0.4.2, the training time took 2.25 min:

However, after the Lightning Migration, the training time took 10 min, or 4x longer:

I used the same machine (Linux on AWS SageMaker Studio Labs) to run both, one after the other. Please advise.

Thanks.

[karl-richter]

@Kevin-Chen0 thanks for raising this, I am having a look!