Add optimum Intel® pass for converting a Huggingface Model to an OpenVINO Model by Anirudh-Swaminathan · Pull Request #1759 · microsoft/Olive

Anirudh-Swaminathan · 2025-04-17T07:29:01Z

Describe your changes

The changes in this PR are for creating an optimum Intel® pass for converting a Huggingface Model to an OpenVINO model, along with the ability to perform weight compression or quantization if necessary.
The aim is to add in the functionality offered by the optimum-cli export openvino command within Olive.

The changes include the addition of a new pass, unit tests for the pass, a few updates to the other OpenVINO passes, and updates to the documentation files as well.

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.
Is this PR including examples changes? If yes, please remember to update example documentation in a follow-up PR.

(Optional) Issue link

Add OpenVINO Optimum Conversion support and update dependencies

Improve warning messages for library name specification and dataset requirement in OpenVINO conversion pass

Improve warning message for unspecified library in OpenVINOOptimumConversion

Add documentation for OpenVINO Optimum Conversion and update version requirements. Also update pass.rst to add in the new pass.

Update test requirements and add unit test for OpenVINO Optimum Conversion pass

Enhance OpenVINOOptimumConversion to handle tokenizer output and improve unit tests

Links now point to pass.rst and to the correct subsection.

…INO passes Also fix device mapping in Quantization for GPU, which had incorrectly been mapped to CPU so far.

Refactor file handling for GenAI models in OpenVINO passes and update tokenizer model paths

…VINO passes

jambayk · 2025-04-17T18:05:18Z

Please rebase or merge main to fix the conflicts that came after #1754 was merged. thanks!

Anirudh-Swaminathan · 2025-04-17T18:14:50Z

Please rebase or merge main to fix the conflicts that came after #1754 was merged. thanks!

Thanks for the heads up! Rebased on top of latest main.

docs/source/reference/pass.rst

Update formatting in pass.rst for OpenVINOOptimumConversion section

docs/source/features/ihv-integration/openvino.md

olive/passes/openvino/optimum_intel.py

Update OpenVINO documentation links to point to the latest reference URLs and improve error handling in OpenVINOOptimumConversion for missing Diffusers package.

Add enumerations for OpenVINO quantization modes, libraries, frameworks, and weight formats

olive/passes/openvino/optimum_intel.py

xiaoyu-work · 2025-04-18T01:08:20Z

Good to thanks for your change!

Move nncf import to validate_config() to avoid doc build errors. Add try-except blocks for nncf and optimum imports in OpenVINO conversion.

xiaoyu-work · 2025-04-18T03:59:47Z

Format check failed. Can you format the code?

Fix an if condition check for quantization parameters provided without providing quant_mode. Avoid access using _value2member_map_ protected member for enum. Instead use list comprehension to iterate through all enum values.

Anirudh-Swaminathan · 2025-04-18T17:29:14Z

Format check failed. Can you format the code?

Thanks! The linter warned about _value2member_map_ being a protected member, so had to refactor the enum access via a list comprehension instead now in my latest commit.

Anirudh-Swaminathan added 11 commits April 17, 2025 10:41

Initial OpenVINO Optimum Conversion pass.

8cc4352

Add OpenVINO Optimum Conversion support and update dependencies

Fix additional linter issues

e83c8ee

Improve warning messages for library name specification and dataset requirement in OpenVINO conversion pass

Minor whitespace fixes

ee25dd4

Improve warning messages for library name specification and dataset requirement in OpenVINO conversion pass

Minor error message fix

33bd5fa

Improve warning message for unspecified library in OpenVINOOptimumConversion

Update OpenVINOOptimumConversion documentation.

411b6f4

Add documentation for OpenVINO Optimum Conversion and update version requirements. Also update pass.rst to add in the new pass.

Initial unit test commit for optimum pass.

2aeedb6

Update test requirements and add unit test for OpenVINO Optimum Conversion pass

Handle tokenizer models in Olive pass.

9344df4

Enhance OpenVINOOptimumConversion to handle tokenizer output and improve unit tests

Fix typos and update links in documentation for OpenVINO passes

bb4efee

Links now point to pass.rst and to the correct subsection.

Add support for copying JSON and text files for GenAI models in OpenV…

a55e102

…INO passes Also fix device mapping in Quantization for GPU, which had incorrectly been mapped to CPU so far.

Correctly handle composite models in optimum intel conversion.

4ae7834

Refactor file handling for GenAI models in OpenVINO passes and update tokenizer model paths

Add support for copying tokenizer and detokenizer directories in Open…

0a2b14b

…VINO passes

Anirudh-Swaminathan force-pushed the user/aswamin2/optimum_intel branch from caa4688 to 0a2b14b Compare April 17, 2025 18:07

jambayk reviewed Apr 17, 2025

View reviewed changes

docs/source/reference/pass.rst Outdated Show resolved Hide resolved

Make doc underline same length as name

dbbd859

Update formatting in pass.rst for OpenVINOOptimumConversion section