Arm_inductor_quantizer for Pt2e quantization

[choudhary-devang]

Title: Enable PyTorch 2 Export Quantization path for ARM CPUs.

Description:

• This PR extends the PyTorch 2 Export Quantization (PT2E Quantization) workflow—originally available only on x86 CPUs—to support ARM platforms. PT2E Quantization is an automated, full-graph quantization solution in PyTorch that improves on Eager Mode Quantization by adding support for functionals and automating the overall process. It is part of the torch.ao module and fully supports quantization when using the compile mode.

Key Changes:

• Introduces ARM-specific support by leveraging oneDNN kernels for matmuls and convolution.

• Integrates pre-defined configuration selection to automatically choose the best quantization settings based on the selected quantization method.

Provides customization options via two flags:

• qat_state: Indicates whether to use Quantization Aware Training (if set to True) or Post Training Quantization (if set to False). The default remains False.
• dynamic_state: Selects between dynamic quantization (if True) and static quantization (if False). The default is also set to False.
  

These options allow users to tailor the quantization process for their specific workload requirements (e.g., using QAT for fine-tuning or PTQ for calibration-based quantization).

Testing and Validation:

The new ARM flow has been thoroughly tested across a range of models with all combinations:
NLP: Models such as BERT and T5.
Vision: Models like ResNet and ViT.
Custom Models: user defined models with various operators.

example script:

import torch
from transformers import BertModel
import copy
import time
from torchao.quantization.pt2e.quantize_pt2e import prepare_pt2e, convert_pt2e
import torchao.quantization.pt2e.quantizer.arm_inductor_quantizer as aiq
from torchao.quantization.pt2e.quantizer.arm_inductor_quantizer import ArmInductorQuantizer
import torch.profiler
import torch._inductor.config as config
# Enable C++ wrapper for Inductor
config.cpp_wrapper = True
config.freezing=True

model_name = "bert-base-uncased"
model = BertModel.from_pretrained(model_name)

# Set the model to eval mode
model = model.eval()

# Create the data, using dummy data here as an example
traced_bs = 32
seq_length = 128
x = torch.randint(0, 10000, (traced_bs, seq_length))
attention_mask = torch.ones((traced_bs, seq_length))
example_inputs = (x, attention_mask)

# Capture the FX Graph to be quantized
with torch.no_grad():
    exported_model = torch.export.export_for_training(model, example_inputs).module()

    # Set up the quantizer and prepare the model for post-training quantization
    quantizer = ArmInductorQuantizer()
    quantizer.set_global(aiq.get_default_arm_inductor_quantization_config(is_dynamic=True))
    prepared_model = prepare_pt2e(exported_model, quantizer)

    # Run the prepared model to apply the quantization
    prepared_model(*example_inputs)

    # Convert the model to the quantized version
    converted_model = convert_pt2e(prepared_model)
    optimized_model = torch.compile(converted_model)
    st = time.time()
    optimized_model(*example_inputs)
    et = time.time()
    print(f"Average time required for inference = {et-st}\n")

cc: @jerryzh168, @fadara01, @Xia-Weiwen

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2139

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure

As of commit 12ecd5a with merge base c9b9adc ():

NEW FAILURE - The following job has failed:

• Run Regression Tests / test-nightly (CUDA Nightly, linux.g5.12xlarge.nvidia.gpu, --pre torch --index-url https://downloa... / linux-job (gh)
  test/integration/test_integration.py::TestUtils::test_get_model_size_autoquant_5_cuda

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

[facebook-github-bot]

Hi @choudhary-devang!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[choudhary-devang]

This pr is moved from torch to torchao due to the Pt2e migration to torchao

old pr link
(pytorch/pytorch#146690)

[choudhary-devang]

Hi @jerryzh168, can you please add the required labels to trigger the pipelines.

[jerryzh168]

please fix ruff errors, you can install pre-commit so it can run automatically when you add a new commit

[jerryzh168]

nit: arm_inductor

[jerryzh168]

for ruff, we are using 0.11.6 and you can run ruff format locally to fix the issue

[maajidkhann]

for ruff, we are using 0.11.6 and you can run ruff format locally to fix the issue

@jerryzh168 Looks like Devang has pushed the fix. Can you now trigger the pipelines again.

[jerryzh168]

done, I also just added @choudhary-devang as a contributor

[jerryzh168]

btw, for inductor fusion passes, you can add to torchao/quantization/pt2e/inductor_passes/ folder for now

[fadara01]

How will this be tested in CI?
We need to make sure that intel changing X86InductorQuantizer (which we inherit from), won't break our functionality

[choudhary-devang]

btw, for inductor fusion passes, you can add to torchao/quantization/pt2e/inductor_passes/ folder for now

Hi @jerryzh168, thanks for adding me as a contributor, and sure i will add the lowering changes for weight as per_tensor in this path in my next pr.

[choudhary-devang]

How will this be tested in CI? We need to make sure that intel changing X86InductorQuantizer (which we inherit from), won't break our functionality

I think currently this isn't covered in CI, Once the test file (https://github.com/pytorch/ao/pull/2139/files#diff-af6ea96fb55af02d2e1c7837c198f7d73a38d542b5ea7000e683422a6b7ca17b) is made part of the CI execution. Then, any change to X86InductorQuantizer will trigger test failures if it breaks ArmInductorQuantizer.

These test files both from x86 and ARM might need to be added to CI execution later on.

[maajidkhann]

Looks like the CUDA nightly pipeline failing is unrelated to this PR and is failing in all other open PR's as well.