[Bug]: AutoDeploy: torch-compile: fix fake ops for prepare_metadata and other ops

[lucaslie]

System Info

Fake ops for our prepare_metadata ops are not correct since the shapes are dependent on the input values of the tensors. This will cause torch-compile to fail.

There is two potential ways we can consider fixing it:

1. Hack to see if we can trick the torch-compile shape analysis
2. Properly declare additional dynamic shapes that will correctly describe the metadata

This should probably also tie into our discussion of how and if we want to use torch-compile at all moving forward...

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

n/a

Expected behavior

n/a

actual behavior

n/a

additional notes

n/a

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

[lucaslie]

Note that fully enabling logit gathering is currently blocked by this ticket