Why do we need to unfreeze the learner everytime before retarining even if learn.fit_one_cycle() works fine without learn.unfreeze()

Part 1 (2019)
15

Thanks for this summary!
Regarding point 3.a:

use lr_find() before fit_one_cycle() to get best suited learning rate for underlying data.

I thought the LR finding occurs after the first fit_one_cycle:

56
561030×260 23.8 KB

Source: Universal Language Model Fine-tuning for Text Classification