Bug: gradient_accumulate_contexts function

[spnova12]

With ref to :

stylegan2-pytorch/stylegan2_pytorch/stylegan2_pytorch.py

Lines 215 to 226 in 0588213

	def gradient_accumulate_contexts(gradient_accumulate_every, is_ddp, ddps):
	if is_ddp:
	num_no_syncs = gradient_accumulate_every - 1
	head = [combine_contexts(map(lambda ddp: ddp.no_sync, ddps))] * num_no_syncs
	tail = [null_context]
	contexts = head + tail
	else:
	contexts = [null_context] * gradient_accumulate_every
	for context in contexts:
	with context():
	yield

I think "map(lambda ddp: ddp.no_sync, ddps)" should be changed to "list(map(lambda ddp: ddp.no_sync, ddps))"
Otherwise, since map is an iterator, it can be used only once, and then the function of gradient_accumulate cannot work well.

[lucidrains]

it is iterated over within the combine contexts function