Time series forecasting using pytorch

Has anybody implemented time series forecasting using pytorch_forecasting.models.nhits or TFT ?
I have two issues

  1. prediction = best_tft.predict(val_dataloader, return_x=True,return_y=True)
    how do i interprete the preidction?
    I want to convert it back to pandas dataframe.

  2. for pytorch_forecasting.models.nhits
    I am getting error
    mat1 and mat2 shapes cannot be multiplied (1x18516 and 24036x64)
    I am not getting how to resolve this?