Question in torch_utils.py

[VvvvvGH]

I found something wired when I am looking at torch_utils.py.
If you want to reduce randomness, shouldn't do as pytorch document suggested to set torch.backends.cudnn.deterministic = True and
torch.backends.cudnn.benchmark = False ?

yolov5/utils/torch_utils.py

Lines 12 to 18 in 7c2832c

	def init_seeds(seed=0):
	torch.manual_seed(seed)
	# Reduce randomness (may be slower on Tesla GPUs) # https://pytorch.org/docs/stable/notes/randomness.html
	if seed == 0:
	cudnn.deterministic = False
	cudnn.benchmark = True

[glenn-jocher]

@VvvvvGH yes you are correct. The comment is incorrect, the settings are coded for the fastest training.

[glenn-jocher]

This should be resolved in 22d6088 now (incorrectly labelled as a #17 fix).

Default training seed changed to 1 now as part of changes.