[Test] Save local model path in PEFT adapter config

[ebsmothers]

It was pointed out in #2025 that the one-liner AutoModelForCausalLM.from_pretrained(model_id) doesn't work for our models trained with LoRA (instead you need to call AutoModelForCausalLM.from_pretrained on a hub model followed by PeftModel.from_pretrained on the LoRA-trained model). This is a demo of saving the checkpointer's directory as part of our adapter_config.json file which should enable this to work.

Edit: after discussing with @pbontrager this won't integrate well with e.g. push to hub since it assumes a local path. So I'm not sure we want to actually land this as is, the better solution may be to make some changes to tune download.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/2026

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• [DomainsOnly] Jobs fail with GLIBC version not found

✅ No Failures

As of commit c69a4f8 with merge base bce7091 ():
💚 Looks good so far! There are no failures yet. 💚

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