benchmark
benchmark copied to clipboard
Correctness test for train()
Can model it after https://github.com/pytorch/torchdynamo/blob/f11a33107407c51e93cba96e26c2f65e56edbc1d/torchdynamo/testing.py#L37