Need updated guide on how to use fast ai in kaggle kernels

I tried to train a model in a kaggle kernel but since input is read only i am not able to run lr_find() and it throws this error:

how do i work around this?
Previously we could specify the temp path as per this link:
but it is not possible in cnn_learner
So what should i do?

Change model_dir parameter when you create your learner to something like ‘/kaggle/working’ as default location is where datasets go and that is read only

thanks a lot

that worked but now when I try to train I get this:
Runtime error:
expected backend CUDA and dtype Float but got backend CUDA and dtype Long
should change the data in my df to float or something?
thank you in advance

hard to know without the code but you can try .astype(‘float’)

Kaggle’s read-only restriction is a bit annoying at times. I go around it by setting learn.path = Path('/') after the learner has been defined. The root folder does not have read-only restrictions.