RuntimeError: CUDA out of memory. Tried to allocate 1.07 GiB (GPU 0; 4.00 GiB total capacity; 2.57 GiB already allocated; 84.45 MiB free; 2.59 GiB reserved in total by PyTorch)

Question

RuntimeError: CUDA out of memory. Tried to allocate 1.07 GiB (GPU 0; 4.00 GiB total capacity; 2.57 GiB already allocated; 84.45 MiB free; 2.59 GiB reserved in total by PyTorch)

HadhamiRjiba opened this issue 3 years ago · 5 comments

Hello,
when i am running "python train.py ",, i have this message in below: " **training, momentum, eps, torch.backends.cudnn.enabled

**RuntimeError: CUDA out of memory. Tried to allocate 1.07 GiB (GPU 0; 4.00 GiB total capacity; 2.57 GiB already allocated; 84.45 MiB free; 2.59 GiB reserved in total by PyTorch)****"

Any idea what might cause this ?

Answer 1 · 2021-06-18T09:06:49.000Z

Is there any way to know how big a model or a network my system can handle
without running into this issue?

Answer 2 · 2021-06-18T11:04:54.000Z

Your GPU ran out of memory. Try lowering the batch size.

Answer 3 · 2021-06-19T07:42:07.000Z

yes!. I reduced the batch size to 1 and now it works with train.py.. but the problem now is in running eval.py (is showing the same error) :
"RuntimeError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 4.00 GiB total capacity; 2.49 GiB already allocated; 44.45 MiB free; 2.57 GiB reserved in total by PyTorch)""..
have you any idea ??

Answer 4 · 2021-06-19T17:51:08.000Z

Hmm. The batch size is 1 by default in eval.py. You could try lowering seq_length. Looks like you only have 4 GB of GPU memory. I used a GPU with 12 GB memory.

Answer 5 · 2021-06-21T09:41:28.000Z

@wmcnally it works..thank you