How can I use the lr_find with K_fold.py example?

[Norman-GM]

As show in the figure,

It raise the overflow problem and loss in nan.
I have tried the lr callback and trainer.tune(model,module) - tuner.lr_find(model, datamodule=module,early_stop_threshold=None)
as shown in the issue of #4846 but it doesn't work.
Seems has nested loop in K-fold loops that cause the lr_find Wrong iteration, which K times the value of num_training, which at least casue the progress output problem!
But I haven't idea to solve it, plz help!

Thanks