Copy and adapt pt2e quantization code to torchao

[jerryzh168]

Summary:
First step of dev-discuss.pytorch.org/t/torch-ao-quantization-migration-plan/2810

core logic of pt2e are duplicated, and also ported test here

previous unfinished migration: #1916

Next:

move meta internal callsites to depend on torchao
Docs will be migrated separately

Test Plan:
pytest test/quantization/pt2e_flow

Reviewers:

Subscribers:

Tasks:

Tags:

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2048

• 📄 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 0050dce with merge base 31f119e ():
💚 Looks good so far! There are no failures yet. 💚

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

[andrewor14]

Hi Jerry, can we add a limited set of tests to show that PT2E in pytorch core vs torchao are numerically equivalent, so users will have more confidence they can migrate without any changes in numerics? We can do this in a follow up PR since this one is already 25k lines

[andrewor14]

Another question is how do we ensure both copies of PT2E remain the same? E.g. if we fix a bug in PyTorch core PT2E, is there an easy way to remind us to fix the same bug in torchao's PT2E? (and vice versa?)

[jerryzh168]

Another question is how do we ensure both copies of PT2E remain the same? E.g. if we fix a bug in PyTorch core PT2E, is there an easy way to remind us to fix the same bug in torchao's PT2E? (and vice versa?)

we want to deprecate the impl in pytorch core soon, and don't want to do updates on both places

[jerryzh168]

Hi Jerry, can we add a limited set of tests to show that PT2E in pytorch core vs torchao are numerically equivalent, so users will have more confidence they can migrate without any changes in numerics? We can do this in a follow up PR since this one is already 25k lines

sure, can do this a bit later