Move quant ops to utils.py

[jerryzh168]

Summary:
We had a lot of "quant primitive" ops that can be expressed with more primitive ops, so these ops are more of a helper functions now, so we moved them to torchao.quantization.utils

we should be able to further deprecate some of the ops after we deprecate subclasses and refactor smoothquant etc. in the future

Also moved TORCH_VERSION_AFTER_{2_2/2_3/2_4} from torchao.quantization.utils to torchao.utils

After the move we have the following ops in quant_primitives.py:

    "safe_int_mm",
    "int_scaled_matmul",
    "choose_qparams_affine",
    "quantize_affine",
    "dequantize_affine",

Test Plan:
python test/integration/test_integration.py
python test/quantization/test_quant_api.py
python test/quantization/test_quant_primitives.py
python test/quantization/test_qat.py

Reviewers:

Subscribers:

Tasks:

Tags:

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/331

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Rebase your PRs: Unstable CUDA signal in CI caused by cudnn 9 update

✅ No Failures

As of commit eebb7a2 with merge base 000a0fd ():
💚 Looks good so far! There are no failures yet. 💚

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

[msaroufim]

I'm having trouble figuring out what should be in quant_primitives vs utils

[jerryzh168]

functions in utils are mostly helper functions that calls the quant_primitive ops with some fixed parameters, e.g.

quantize_affine can support: symmetric/asymmetric, per tensor/group/channel etc., int8/int4/int3

helper function in utils can be: int8_symmetric_per_tensor_quant that calls quantize_affine op with fixed settings