Add lr_find before training, it changes the training result, why is it


(this one added shuffle_train= False to make sure the dataloader is not shuffled)

As shown in the screenshot, insert a lr_find in front of the fit, it will change the training result.

Is this suppose to happen? Is this a bug or somthing that I missed?

What version of fastai are you using? This was a known bug that was fixed recently, but has not been tested. If you’re on the latest version and are facing this, it would be great if you could provide a minimal reproducible example and post on this GitHub issue:

1 Like

I’m on the lastest version v2.1.5. yep. I will post it on gibhub.