Save graph prints as lazy strings instead of eager

[jamesjwu]

Stack from ghstack (oldest at bottom):

• -> Save graph prints as lazy strings instead of eager #137700

My previous PR made these eager instead of lazy, making it regress some small pr time benchmarks that don't use structured trace. Making them lazy again by putting them under a function call fixes the issue. Production jobs should not be affected by the change either way, because structured logs are turned on and evaluated there.

Without this diff:

collecting compile time instruction count for aotdispatcher_inference_nosubclass_cpu
compile time instruction count for iteration 0 is 10241330219
compile time instruction count for iteration 1 is 1970239415
compile time instruction count for iteration 2 is 1955721527
compile time instruction count for iteration 3 is 1949317363
compile time instruction count for iteration 4 is 1947567621
collecting compile time instruction count for aotdispatcher_training_nosubclass_cpu
compile time instruction count for iteration 0 is 4203676200
compile time instruction count for iteration 1 is 3739626299
compile time instruction count for iteration 2 is 3736479815
compile time instruction count for iteration 3 is 3734131630
compile time instruction count for iteration 4 is 3733066773
collecting compile time instruction count for aotdispatcher_inference_subclass_cpu
compile time instruction count for iteration 0 is 5621749132
compile time instruction count for iteration 1 is 5621658470
compile time instruction count for iteration 2 is 5620004680
compile time instruction count for iteration 3 is 5617585817
compile time instruction count for iteration 4 is 5618850646
collecting compile time instruction count for aotdispatcher_training_subclass_cpu
compile time instruction count for iteration 0 is 10005741891
compile time instruction count for iteration 1 is 10003433898
compile time instruction count for iteration 2 is 10002792031
compile time instruction count for iteration 3 is 10000217081
compile time instruction count for iteration 4 is 10000992803

With this diff:

collecting compile time instruction count for aotdispatcher_inference_nosubclass_cpu
compile time instruction count for iteration 0 is 10128746827
compile time instruction count for iteration 1 is 1965846584
compile time instruction count for iteration 2 is 1952468717
compile time instruction count for iteration 3 is 1946109980
compile time instruction count for iteration 4 is 1944459203
collecting compile time instruction count for aotdispatcher_training_nosubclass_cpu
compile time instruction count for iteration 0 is 4052066041
compile time instruction count for iteration 1 is 3589558675
compile time instruction count for iteration 2 is 3585668104
compile time instruction count for iteration 3 is 3584020963
compile time instruction count for iteration 4 is 3583162039
collecting compile time instruction count for aotdispatcher_inference_subclass_cpu
compile time instruction count for iteration 0 is 5617038583
compile time instruction count for iteration 1 is 5617303182
compile time instruction count for iteration 2 is 5613289422
compile time instruction count for iteration 3 is 5612847744
compile time instruction count for iteration 4 is 5612093362
collecting compile time instruction count for aotdispatcher_training_subclass_cpu
compile time instruction count for iteration 0 is 9705752209
compile time instruction count for iteration 1 is 9704758787
compile time instruction count for iteration 2 is 9703760884
compile time instruction count for iteration 3 is 9704687557
compile time instruction count for iteration 4 is 9704824140

Specifically, aotdispatcher_training_nosubclass_cpu and aotdispatcher_training_subclass_cpu look better

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137700

• 📄 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.

❌ 1 New Failure

As of commit 1915f6f with merge base 839d356 ():

NEW FAILURE - The following job has failed:

• Lint / lintrunner-noclang / linux-job (gh)
  >>> Lint for torch/_functorch/_aot_autograd/jit_compile_runtime_wrappers.py:

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

[bdhirsh]

we have a LazyString class we use in a few other places to lazily materialize strings for logging - maybe we should use it here for consistency (to be fair it is a pretty tiny wrapper) https://github.com/pytorch/pytorch/blob/main/torch/_logging/_internal.py#L1083

[jamesjwu]

Yeah in this case since we're passing a lambda anyway I figured making it lazy string was literally more overhead

[Skylion007]

You could technically just pass a functools.partial :P

[jamesjwu]

@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 mandatory check(s) failed. The first few are:

• Lint / lintrunner-noclang / linux-job

Dig deeper by viewing the failures on hud

Details for Dev Infra team

Raised by workflow job

Failing merge rule: Core Maintainers

[laithsakka]

if those effect
"basic_modules_ListOfLinears_inductor": 3.7966236673926,
"basic_modules_ListOfLinears_inductor_gpu_force_shape_pad": 2.1187519601234
you might have to rebase an update expected value if you are not already on top of:
#137541

[github-actions]

Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as Stale.
Feel free to remove the Stale label if you feel this was a mistake.
If you are unable to remove the Stale label please contact a maintainer in order to do so.
If you want the bot to never mark this PR stale again, add the no-stale label.
Stale pull requests will automatically be closed after 30 days of inactivity.