Upgrading PyTorch to v1.11.0 and torchvision to v0.12.0 and torchaudio to v0.11.0

[atalman]

Hello,

We released

pytorch v1.11.0
torchvision v0.12.0
torchaudio v0.11.0
The wheel links are:

pytorch

CUDA 11.3

download.pytorch.org/whl/cu113/torch-1.11.0%2Bcu113-cp37-cp37m-linux_x86_64.whl

torchvision

CUDA 11.3
download.pytorch.org/whl/cu113/torchvision-0.12.0%2Bcu113-cp37-cp37m-linux_x86_64.whl

torchaudio

CUDA 11.3
download.pytorch.org/whl/cu115/torchaudio-0.11.0%2Bcu115-cp37-cp37m-linux_x86_64.whl

Other notes
If you require wheels for Python 3.8 / 3.9 or 3.10, you can find the links here: download.pytorch.org/whl/torch_stable.html

We're looking to having it updated in Colab.

Thanks very much.

cc'ing @colaboratory-team

Similar to #2372

[craigcitro]

Thanks for the note!

@atalman is there info about your expected CUDA compatibility published somewhere? I ask because we're currently on CUDA 11.1, and updating is on our radar but going to require significant work. The download page only shows CUDA 11.3 and 11.5; we may be blocked on upgrading quickly if those are the only versions available.

[atalman]

Hi @craigcitro as per previous story : #2372 CUDA 11.3 should be working, CUDA have something called "enhanced-backward-compatibility", which means that CUDA-11.X releases should work with any CUDA-11.0 compatible driver.
We run small test of our cuda 11.3 binaries in colab and it worked.

cc'ing: @malfet @seemethere

[atalman]

Any update on this issue @craigcitro ?

[craigcitro]

Hi @atalman -- I'm taking a look. (I'm getting tripped up by python package version conflicts, which I don't think are actually related to pytorch, but it may take some doing to untangle.)

There are also #2649 and #2656 as possible speed bumps.

[malfet]

Hi @craigcitro , fyi in this collab I've tested that CUDA-11.3 pip wheels from PyTorch-1.10.2 just work out of box. I assume it should be the same for PyTorch-1.11 and CUDA-11.3

[craigcitro]

This should be fixed!