Hi. I am experimenting with the Nixtla toolbox to (at least maybe) finally replace my home-grown time series prediction libraries. I like it a lot so far. It seems though, if you walk through tutorials like https://nixtla.github.io/mlforecast/docs/tutorials/electricity_load_forecasting.html that if you fit an MLForecast object, that the actual model objects inside the "Models" dictionary loose their information about their fitted parameters. LIke the coefficients in a OLS regression. And lin_reg.coef_ will lead to an exception, since the object is still the unfitted sklearn object. I was too lazy to debug it, but you are cloning the models in the fit loop.