Hey guys, it looks like the model having problem when passing more than one gpu id.
It crashes after trying to fetch the model's generator, as the DataParallel object doesn't have one (It only has the module object)
It looks like this breaks training not only in this specific line.
Any advice?
(In order to reproduce the problem I'm facing just pass -guid 0 1)
how to solve this question?
the same problem...
Could you post the trackback ? Is it the newest code?
Yeah. Sorry, multigpu is actually not supported at the moment. We are waiting on some work from the pytorch team. Currently even if you get it to run, it is not much faster.
Most helpful comment
Yeah. Sorry, multigpu is actually not supported at the moment. We are waiting on some work from the pytorch team. Currently even if you get it to run, it is not much faster.