Add a configuration for training Llama 2 13B QLoRA on single device

[rohan-varma]

Context

• We can finetune 13B on a single device w/QLoRA, but didn't have an explicit config for doing so. This PR adds that.
• Also added the finetuning option in the output of tune ls.
• Want to get this in so users can OOTB finetune w/13B, so doesn't have CI yet. Filed Add CI for 13B QLoRA #780 to add CI.

Changelog

• ...

Test plan

• Loss looks reasonable:
  
• Will run eval as well
• tune ls output:

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/779

• 📄 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 cc8682d with merge base da608e6 ():
💚 Looks good so far! There are no failures yet. 💚

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