m.sched.plot() is also showing a blank chart And the loss is not decreasing.
@radek.
Are you able to successfully run a fit loop otherwise?
How many batches are in your data? Sometime the learning rate finder struggles with data sets that have a small number of batches.
Check how much memory you are using: nvidia-smi
Check if that happens when you set the batch size to 1.
i tried batch size of 512,128,64,8,1 but it didnt work and have total data points around 600000
The image is quite small, so I didn’t see if you used params. You may try to pass a very broad range and see if that helps:
learn.lr_find(1e-7,1e2)
learn.sched.plot(n_skip=0, n_skip_end=0)