I have some problems with my code that runs fine on my computer but not on FloydHub CPU (pytorch-0.4)
Are you working with the same PyTorch version that you are using locally?
the CPU utilization goes to 0% and the programs simply stops, the forward() pass is never complected. The same function works if I call it from the main process, but not when it's called from the object run() function.
No error is reported and the job continues to run, consuming cpu time but without doing anything.
From my limited understanding, this seems the classical deadlock behaviour. Did you check the best practice section when using multiprocessing before?
Hope that helps.