Optional: use inv_scale for WeightTensorWithLinearActivationScaleMetadata

[Xia-Weiwen]

Summary
Use inv_scale on CPU to avoid division at runtime to improve performance. Used for SmoothQuant.

Test plan
pytest -sv test/prototype/test_smoothquant.py

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3176

• 📄 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 c7e242c with merge base ff16308 ():
💚 Looks good so far! There are no failures yet. 💚

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

[Xia-Weiwen]

Hi @jerryzh168 Can you please review this PR? Thanks.

[jerryzh168]

we are deprecating these utils though to_weight_tensor_with_linear_activation_scale_metadata

[jerryzh168]

we are deprecating these utils though to_weight_tensor_with_linear_activation_scale_metadata

[Xia-Weiwen]

we are deprecating these utils though to_weight_tensor_with_linear_activation_scale_metadata

Hi @jerryzh168 Thanks for reviewing. This PR modifies the to_weight_tensor_with_linear_activation_scale_metadata path, which is defined here:

ao/torchao/quantization/linear_activation_scale.py

Lines 118 to 120 in 7e68d5e

	to_weight_tensor_with_linear_activation_scale_metadata = (
	WeightTensorWithLinearActivationScaleMetadata.from_float
	)

And this PR adds a new argument use_inv_scale to WeightTensorWithLinearActivationScaleMetadata.from_float:

ao/torchao/quantization/linear_activation_scale.py

Lines 73 to 78 in 7e68d5e

	def from_float(
	cls,
	input_float: torch.Tensor,
	scale: torch.Tensor,
	):
	return cls(input_float, scale)

and modifies the implementation of the WeightTensorWithLinearActivationScaleMetadata.

Can you please review again? Thanks.