typing proxy_tensor.py

[aorenste]

Stack from ghstack (oldest at bottom):

• WIP: fake tensor SymInt support #127596
• fake_tensor: flatten cache key #129780
• typing fake_tensor.py #128041
• -> typing proxy_tensor.py #129182
• Fix test_public_bindings to skip examples #129524

cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin @XilunWu @wanchaol @fduwjj @wz337 @tianyu-l @wconstab @yf225 @chauhang @d4l3k @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/129182

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 5 Unrelated Failures

As of commit 845f3f1 with merge base 9ae78a5 ():

NEW FAILURES - The following jobs have failed:

• BC Lint / bc_linter (gh)
  Process completed with exit code 1.
• inductor-periodic / cuda12.1-py3.10-gcc9-sm80 / test (inductor_torchbench_smoketest_perf, 1, 1, linux.gcp.a100) (gh)
  moco

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (aot_eager_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (similar failure)
  moco
• inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (dynamic_aot_eager_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (similar failure)
  moco
• inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (dynamo_eager_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (similar failure)
  moco

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• inductor / cuda12.1-py3.10-gcc9-sm86 / test (dynamic_inductor_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (trunk failure)
  moco
• inductor / cuda12.1-py3.10-gcc9-sm86 / test (inductor_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (trunk failure)
  moco

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Skylion007]

is this safe? One checks for inheritance, the other checks for protocol?

[aorenste]

Actually - I took another look. This is correct. The issue is that is_traceable_wrapper_subclass() only returns true for isinstance(t, Tensor) - so since it's an 'or' you only need the isinstance(t, torch.Tensor).

[Skylion007]

I guess I'm confused about what the semenatic are for is_traceable_wrapper_subclass and if it will ever have subclasses that are not subclassing Tensor?

[aorenste]

The doc in is_traceable_wrapper_subclass says:

Returns whether or not a tensor subclass that implements __torch_dispatch__
    is 'traceable' with torch.compile.

which (to me) implies that it will always be a tensor subclass.

[eellison]

@Chillee would you take this one ?

[Chillee]

@eellison I'll take a look.