Release 2.2.2 validations

[atalman]

🐛 Describe the bug

PyTorch-2.2.2 we need to make sure that:

• [ ] CUDA
  • [x] pypi binaries with slimmed dependencies are usable in standard AWS containers (amazonlinux:2 regression in 1.13)
  • [x] pypi binaries with slimmed dependencies are usable with stock Ubuntu-20.04: https://github.com/pytorch/pytorch/issues/91067 . Test: https://github.com/pytorch/builder/actions/runs/7973707143/job/21768131794#step:11:5171
  • [ ] PyTorch+CUDA11.8 and CUDA 12.1 has a working FFT on 4090
  • [x] Check cuda 1.12.1 update issue: https://github.com/pytorch/pytorch/issues/94772 with small wheels
  • [ ] torch.compile
  • [x] Basic test works (for example see test mentioned in https://github.com/openai/triton/pull/1176 ) in PyTorch docker container
  • [ ] torch.compile produces a binary which can be used on 3090
  • [x] torch.compile raises an error if used on Windows. Test: https://github.com/pytorch/builder/actions/runs/7973707143/job/21768117984#step:9:534
  • [x] torch.compile works on 3.11 : Test: https://github.com/pytorch/builder/actions/runs/7973707143/job/21768126221#step:11:5565
  • MPS
  • [x] Resnet is usable out of the box (i.e. https://github.com/pytorch/builder/blob/main/test/smoke_test/smoke_test.py passes for MPS device). Test: https://github.com/pytorch/builder/actions/runs/7973707143/job/21768145633#step:9:312
• [ ] Validate docker release builds

Same as: https://github.com/pytorch/pytorch/issues/120250

Versions

2.2.2

[atalman]

Possibility to automate this task: pypi binaries with slimmed dependencies are usable in standard AWS containers (amazonlinux:2 regression in 1.13)

[huydhn]

Validation jobs https://github.com/pytorch/builder/actions/runs/8432086196

[huydhn]

Confirm that import and smoke test works with docker pull amazonlinux:2023 (CPU, no CUDA)