Make QuantizeTensorToFloat8Kwargs a frozen Dataclass to enable torch export

[asfiyab-nvidia]

Regular Dataclasses are not supported by the Dynamo Tracer as they're dispatched to UserDefinedObjectVariable which doesn't have an implementation for as_proxy. However, Frozen Dataclasses are supported by the Dynamo Tracer as the as_proxy method is defined for them (source).

Marking QuantizeTensorToFloat8Kwargs as a frozen dataclass enables torch.export on the Float8Tensor subclass.

Fixes #3928

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4011

• 📄 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 b491570 with merge base 5a5029d ():
💚 Looks good so far! There are no failures yet. 💚

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

[asfiyab-nvidia]

@jcaip can you please help with the review? Thanks

[jcaip]

Thanks for the fix @asfiyab-nvidia. Can we also add an export test to test/quantization/quantize_/workflows/float8/test_float8_tensor.py so that this doesn't happen again?

[asfiyab-nvidia]

@jcaip I added another comment on the issue that motivates this change. Based on the comment, would you still recommend the change the PR proposes to maintain compatibility with PyTorch < 2.7 where unwrap_tensor_subclass is neeeded?