WilkoLu
WilkoLu
@microsoft-github-policy-service agree
> I've added a commit to work out some conflicts. > > @WilkoLu , I hope that's all as you still want. From the comment in the issue, you said...
> The old train code should train fine on multi GPU, I used 4 p100 for the original paper. Recently I have used 4 a100 as well with the new...