Add support for Quark onnx quantization

[gengxinwu]

Describe your changes

What it does:

1. The QuarkQuantization pass can also quantize ONNX format models through olive run interface
2. Support int8/uint8/int16/uint16/int32/uint32/bf16/bfp16 quantization data types
3. Support commonly used algorithms, like CLE, SmoothQuant, AdaRound and AdaQuant

What is next:

1. Add integration with olive quantize interface
2. Integrate higher versions of amd-quark
3. Support more data formats, including MX formats
4. Support more algorithms, like GPTQ, QuaRot ...

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.

(Optional) Issue link

[Copilot]

Pull Request Overview

This PR adds ONNX model quantization support to the Quark quantizer pass. The existing QuarkQuantization pass only supported HuggingFace models; now it supports both ONNX and HuggingFace models.

Key changes:

• Extended QuarkQuantization pass to handle ONNXModelHandler in addition to HfModelHandler
• Added new ONNX-specific quantization logic and configuration preparation utilities
• Included test coverage for the new ONNX quantization functionality

Reviewed Changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
test/requirements-test.txt	Added amd-quark dependency version 0.10
test/passes/quark_quantizer/init.py	Created package initialization file for quark quantizer tests
test/passes/quark_quantizer/test_quark_onnx_quantization.py	Added test case for static QDQ U8S8 quantization
olive/passes/quark_quantizer/quark_quantization.py	Extended pass to support ONNX models with new configuration parameters and _run_quark_onnx method
olive/passes/quark_quantizer/onnx/init.py	Created package initialization file for ONNX quantizer
olive/passes/quark_quantizer/onnx/quantize_quark.py	Implemented ONNX model quantization using Quark's ModelQuantizer
olive/passes/quark_quantizer/onnx/configuration_preparation.py	Added configuration mapping utilities for converting dictionaries to Quark ONNX config objects

[Copilot]

This TODO comment indicates incomplete implementation. The update_algo_config function handles only AdaRoundConfig, AdaQuantConfig, CLEConfig, and SmoothQuantConfig, but the algorithm_mapping dictionary includes GPTQConfig, AutoMixprecisionConfig, and QuarotConfig which are not configured. Consider either implementing the missing algorithm configurations or documenting which algorithms are intentionally not yet supported.

Suggested change

	else:
	# TODO(Gengxin): Configure the rest algorithms
	pass
	elif isinstance(algo_config, GPTQConfig):
	# TODO: Implement configuration for GPTQConfig if/when fields are known
	logger.warning("GPTQConfig configuration is not yet supported. Using default values.")
	elif isinstance(algo_config, AutoMixprecisionConfig):
	# TODO: Implement configuration for AutoMixprecisionConfig if/when fields are known
	logger.warning("AutoMixprecisionConfig configuration is not yet supported. Using default values.")
	elif isinstance(algo_config, QuarotConfig):
	# TODO: Implement configuration for QuarotConfig if/when fields are known
	logger.warning("QuarotConfig configuration is not yet supported. Using default values.")
	else:
	logger.warning("Unknown algorithm config type: %s. No configuration applied.", type(algo_config).__name__)

[Copilot]

The description states 'List of nodes or subgraphs' but the type is declared as dict. This is inconsistent. Either update the description to match the dict type (e.g., 'Dictionary defining nodes or subgraphs excluded from quantization') or change the type to list if it should actually be a list.

Suggested change

	description="List of nodes or subgraphs excluded from quantization. Default is None.",
	description="Dictionary defining nodes or subgraphs excluded from quantization. Default is None.",