Notebook_launcher: Importing libraries initializes CUDA

[NHomer-Edi]

After importing the libraries, CUDA is initialized, which means that running notebook_launcher results in a ValueError.

print(f"{torch.cuda.is_initialized()}")

from fastai.distributed import *
from fastai.vision.all import *
from fastai.vision.models.xresnet import *
from accelerate import notebook_launcher
from accelerate.utils import write_basic_config

print(f"{torch.cuda.is_initialized()}")

Prints

False
True

Which results in the following error, when running notebook_launcher():

ValueError: To launch a multi-GPU training from your notebook, you need to avoid running any instruction using torch.cuda in any cell. Restart your notebook and make sure no cells use any CUDA function.

Documentation (https://docs.fast.ai/tutorial.distributed.html) could be made clearer to explain how/where to import libraries to avoid throwing this error.