skip compiling opt step instead of erroring if opt_in_bwd=True

[felipemello1]

Context

What is the purpose of this PR? Is it to

• add a new feature
• fix a bug
• update tests and/or documentation
• other (please add here)

Changelog: as in the title

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/2827

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 72d0b6f with merge base 2344509 ():
💚 Looks good so far! There are no failures yet. 💚

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

[joecummings]

Why? This doesn't happen that far into the process and it's a simple solution to just change compile optimizer step to false.

If anything, you could provide a better error message like "If you would still like to compile the model and loss, you can specify everything in the config like

blah blah blah"

[felipemello1]

Why? This doesn't happen that far into the process and it's a simple solution to just change compile optimizer step to false.

If anything, you could provide a better error message like "If you would still like to compile the model and loss, you can specify everything in the config like

blah blah blah"

I was doing llama 70B. It takes several minutes for it to start (not sure why). And then it errored. So for me, it felt far into the process. If you dont like skipping and prefer it to just error, i could do this logic in the init instead.