[Thiago Crepaldi (thiagocrepaldi)]

Describe your changes

QuarkQuantization now requires and is tested against amd-quark 0.13, the current Quark release.

Why. The AMD CI installs amd-quark==0.10 (test/requirements-test-amd.txt), three releases behind the API the pass has used since #2532 (amd-quark>=0.12). So the pass's torch path wasn't tested against the Quark it targets.

Impact on users. The released olive-ai 0.13.0 still uses the Quark 0.11 API, and with any amd-quark since 0.12 its torch path fails:

ImportError: cannot import name 'revert_model_patching' from 'quark.torch.utils.llm'

pip install olive-ai amd-quark gives exactly that combination today. Main is fixed (#2532) but not released.

This PR:

• test/requirements-test-amd.txt: amd-quark==0.13.
• QuarkQuantization requires amd-quark>=0.13.0, in four places:
  • the olive_config.json extra;
  • the runtime version gates of the ONNX and torch paths;
  • the tests of those gates;
  • docs/source/features/quantization.md.

Verification. Tested with amd-quark 0.13, onnxruntime 1.28, torch 2.13 and transformers 5.15 / 5.22:

• pytest test/passes/quark_quantizer passes (10 tests).
• The pass runs AWQ uint4_wo_128 with a Hugging Face export on a tiny Llama.
• The same flow runs in AMD Quark's CI on a Windows ROCm GPU (Strix Halo, gfx1151).

Request. Could this go into a release soon? Every released olive-ai fails with the current amd-quark.

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 (not run; the change is version strings only)
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.

QuarkQuantization requires amd-quark >= 0.13.

🤖 Generated with Claude Code

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟢 Approval recommended

No active review issues are reported for the amd-quark 0.13 dependency and test-coverage updates.

No open issues 🎉

What changed in this PR

Aligns QuarkQuantization dependencies and AMD test coverage with amd-quark 0.13, building on the API updates introduced in #2532.

Changes:

• Raises the minimum amd-quark version to 0.13 for both ONNX and PyTorch paths.
• Pins AMD test dependencies to 0.13 and updates version-gate tests to reject 0.12.
• Documents the new minimum version.

File	Description
test/​requirements-test-amd.txt	Pins amd-quark to 0.13 for AMD testing.
test/​passes/​quark_quantizer/​test_quark_onnx_quantization.py	Updates both version-gate tests and expected errors.
olive/​passes/​quark_quantizer/​quark_quantization.py	Requires amd-quark ≥0.13.0 in both runtime paths.
olive/​olive_config.json	Updates the pass dependency to amd-quark ≥0.13.0.
docs/​source/​features/​quantization.md	Documents the amd-quark 0.13 requirement.

_{Review effort: Balanced}

[Thiago Crepaldi (thiagocrepaldi)]

kunal-vaishnavi

[Copilot]

🟢 Approval recommended

No active review issues are identified for the amd-quark 0.13 dependency and test-environment alignment.

No open issues 🎉

_{Review effort: Balanced}

[Baiju Meswani (baijumeswani)]

/azp run Olive CI

[azure-pipelines]

Commenter does not have sufficient privileges for PR 2793 in repo microsoft/Olive

[Thiago Crepaldi (thiagocrepaldi)]

Thanks for the approvals. The only failing check is Docs, and it's GitHub rate-limiting the link check: both broken links are 429 Too Many Requests on microsoft/olive-recipes pages, not links this PR touches, and the same workflow passes on main. Could you rerun Docs (or merge if a rerun hits the limit again)?

[kunal-vaishnavi]

/azp run Olive CI

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Thiago Crepaldi (thiagocrepaldi)]

Docs build passed, but it seems hf_token is missing in the linux cpu job

[kunal-vaishnavi]

/azp run Olive CI

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).