Add legacy note to autograd.profiler doc.

[lucasb-eyer]

Via google search I got to torch.autograd.profiler and implemented my code with it. Only to be taken by surprise finding torch.profile.profiler, which has a note saying the autograd one is legacy.

This just adds such note to autograd.profiler to avoid this confusion and waste of time to future people in my situation.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/157459

• 📄 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.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 989980b with merge base 510c398 ():

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

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

• trunk / macos-py3-arm64 / test (mps, 1, 1, macos-m2-15) (gh) (trunk failure)
  Process completed with exit code 1.

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

[lucasb-eyer]

@pytorchbot label "topic: not user facing"

[sraikund16]

This note is not quite correct. The autograd profiler is the backend that the profiler.profiler uses.

[lucasb-eyer]

I was careful to use the exact wording of the note I found in torch.profiler here, which calls this "legacy".

Also, the torch.profile profiler uses KinetoStepTracker, but not the class profile. I only annotate the doc-comment from the class profile (which is what I mistakenly used), not the whole module.

So, I do think it's actually correct. If you still disagree, any edit you suggest?

[lucasb-eyer]

ping

[sraikund16]

@lucasb-eyer sorry for delay, was out on holiday.

So profiler.profile uses autograd.profile here: https://github.com/pytorch/pytorch/blob/main/torch/profiler/profiler.py#L190
and it is imported here: https://github.com/pytorch/pytorch/blob/main/torch/profiler/profiler.py#L16

The current mechanism is that profiler.profile mostly handles scheduling and some callback registering but autograd.profile is what is actually responsible for calling the pybind functions that will start the profiling. Essentially, the profiler.profile is a wrapper around the autograd implementation.

Sorry if this has caused confusion! We should be more clear in the note to explain this implementation and suggest users to use the profiler.profile as the frontend.

[lucasb-eyer]

No worries. You are right, thanks for pointing to the exact code! I changed the wording to now (hopefully) be correct:

This is the backend, most people should use :mod:torch.profiler instead.

The important part is just to redirect most users there.

[sraikund16]

LGTM!

[lucasb-eyer]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

[pytorchmergebot]

Merge failed

Reason: 1 jobs have failed, first few of them are: linux-binary-manywheel / manywheel-py3_9-rocm6_4-test

Details for Dev Infra team

Raised by workflow job

[lucasb-eyer]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here