PyTorch 1.12.0 for scheduled CI

[ydshieh]

What does this PR do?

After the slack discussion, using PyTorch 1.12.0 for scheduled CI is a better idea, so we can see what to fix.

Another reason is that this (messy) block

transformers/docker/transformers-pytorch-gpu/Dockerfile

Lines 19 to 22 in 39dad97

	RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' || VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
	RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' || VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
	RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' || VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113

will change torch 1.11 (if specified) back to 1.12, as torchvision and torchaudio are installed separately from torch without specifying versions. It's better to avoid such situations.

Regarding the torch/torchvision/torchaudio correspondence, I have better approach in my past-ci PR.

[stas00]

LGTM

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.