Fix all RuntimeErrors during weights_only load from being erroneously reported with the weights_only message

[mikaylagawarecki]

Caught in above PR #127627

Stack from ghstack (oldest at bottom):

• Fix failures when default is flipped for weights_only #127627
• -> Fix all RuntimeErrors during weights_only load from being erroneously reported with the weights_only message #132349

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/132349

• 📄 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.

✅ No Failures

As of commit c7fd26f with merge base 57d1ffc ():
💚 Looks good so far! There are no failures yet. 💚

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

[albanD]

Not sure what's the impact of this change tbh. Is Pickle handling this type of error specially?
Also should we add a new unit test to ensure whatever this is fixing remains fixed?

[mikaylagawarecki]

@albanD main change is that we were always catching any RuntimeError (e.g. those that a corrupted checkpoint would cause) and reporting it with the weights_only message when using weights_only see here and the message here using the UnpicklingError in the unpickler makes sure that we specifically report only weights_only errors with that message and yep sorry will add a test (I think it was tested in the above PR when I set weights_only to True, but then stopped being tested when I flipped the default back to None)

[albanD]

But aren't other things maybe going to raise this type of error as well? Why use UnpicklingError vs WeightsOnlyFailureError ?

[mikaylagawarecki]

Hm not sure how else would an UnpicklingError get raised since we re-implemented the entire Unpickler and don't inherit from pickle.Unpickler, but fair enough, perhaps there's something I missed

[albanD]

Ho that sounds fair then. Feel free to keep it as is then!

[albanD]

SGTM!

[pytorchmergebot]

Rebased gh/mikaylagawarecki/217/orig onto refs/remotes/origin/viable/strict because #127627 was rebased, please pull locally before adding more changes (for example, via ghstack checkout https://github.com/pytorch/pytorch/pull/132349)