medium10 pts · first answer countsgpu-computinggpu-memorypytorch

empty_cache does not help

You call torch.cuda.empty_cache() after each step, but memory still grows until OOM. What is the most likely explanation?

A.empty_cache only works on CPU tensorsB.empty_cache must be called twice per stepC.Live tensors are still referenced, so stay allocatedD.GPU memory can never be released in PyTorch

Discussion 0

Join free to post in the discussion.

empty_cache does not help

Discussion 0

Related questions