Hi everyone,
I was looking at Fastai score in Cifar10 on DawnBench (https://dawn.cs.stanford.edu/benchmark/#cifar10) where the condition to win the “Training Time” category is : Time taken to train an image classification model to a test accuracy of 94% or greater on CIFAR10.
I’d like to reproduce the test using different GPUS and ResNet versions, as to benchmark the combos’ minimal time to reach 94%.
Is there a way to use a Callback in learn.fit_one_cycle()
so that it keeps training "up to 1000 epochs until it achieves accuracy > 0.94
" ?
I tried EarlyStoppingCallback
but this detects when the metric stops improving, not when it reaches a specific value.
Or maybe it doesn’t make sense ?