Closed MichaelMonashev closed 3 years ago
https://github.com/davidtvs/pytorch-lr-finder needs to specify criterion to find best LR:
lr_finder = LRFinder(model, optimizer, criterion, device) lr_finder.range_test(train_loader, end_lr=end_lr, num_iter=30, step_mode="exp", accumulation_steps=accumulation_steps) lr_finder.reset()
How to pass it?
Beyond the scope of this repository. The bench contains criterion and outputs the losses in output as dict keys, you'd likely have to modify the finder to access the loss that way
https://github.com/davidtvs/pytorch-lr-finder needs to specify criterion to find best LR:
How to pass it?