Llama3.2 vision model AOTI integration

[larryliu0820]

🚀 The feature, motivation and pitch

Need work to enable:

python3 torchchat.py export llama3.2-11B --output-aoti-package-path exportedModels/llama3_2_artifacts.pt2

Features to be added:

• Run AOTI export on preprocess, vision encoder and text decoder separately.
  • This requires dependency on torchtune's exportable modules and perform module transformation needed.
• Package them into a single pt2 file.

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

[HonestDeng]

I want to be a long-time contributor. This issue looks very interesting and challenging for me. Can I get any help and solve this issue?