The optimal range of the ’alpha‘ parameter in the TAU model.

[QUSTwh]

When using the TAU model, what value should the 'alpha' parameter in the LOSS function be set to?
What impact does the size of this parameter have?
loss = self.criterion(pred_y, batch_y) + self.hparams.alpha * self.diff_div_reg(pred_y, batch_y)