Skip to content

[LPT] QDQ stripping #32266

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 26 commits into
base: master
Choose a base branch
from
Open

[LPT] QDQ stripping #32266

wants to merge 26 commits into from
+377 −72

Conversation

@v-Golubev
Copy link
Contributor

@v-Golubev v-Golubev commented Oct 1, 2025 •
edited
Loading

Details:

  • This PR Introduces a new transformation pipeline for GPU plugin which enables QDQ stripping for quantized i16 activations

Tickets:

@v-Golubev v-Golubev requested review from a team as code owners October 1, 2025 11:46
@v-Golubev v-Golubev requested review from mryzhov and removed request for a team October 1, 2025 11:46
@github-actions github-actions bot added category: GPU OpenVINO GPU plugin category: transformations OpenVINO Runtime library - Transformations category: LP transformations OpenVINO Low Precision transformations labels Oct 1, 2025
@v-Golubev v-Golubev force-pushed the vg/lpt/qdq_stripping branch from 5f34c30 to e5042b0 Compare October 1, 2025 15:06
@mklimenk mklimenk mentioned this pull request Oct 6, 2025
Draft
@moslex moslex added this to the 2025.4 milestone Oct 27, 2025
@v-Golubev v-Golubev force-pushed the vg/lpt/qdq_stripping branch 2 times, most recently from 1a40855 to a911da0 Compare October 27, 2025 17:46
@v-Golubev v-Golubev force-pushed the vg/lpt/qdq_stripping branch from 1e75852 to 8faa050 Compare October 31, 2025 09:55
@v-Golubev
Copy link
Contributor Author

v-Golubev commented Nov 4, 2025

@aobolensk could you please review LPT part?
@isanghao could you please review GPU plugin and tests part?

aobolensk
aobolensk reviewed Nov 4, 2025
View reviewed changes
@maxnick maxnick added this pull request to the merge queue Nov 5, 2025
isanghao
isanghao reviewed Nov 5, 2025
View reviewed changes
src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp Outdated
manager.register_pass<ov::pass::ConvertQuantizeDequantize>(TypeVector{i16, u16}, TypeVector{f32}, true);
// 3. Strip FQ layers with unsupported levels
bool replace_with_clamp = ov::util::getenv_bool("REPLACE_QDQ_WITH_CLAMP", true);
std::cout << "[ QDQ STRIPPING INFO ] replace_with_clamp = " << replace_with_clamp << std::endl;
Copy link
Contributor

@isanghao isanghao Nov 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please remove unnecessary log message

Copy link
Contributor Author

@v-Golubev v-Golubev Nov 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Removed, thanks. The REPLACE_QDQ_WITH_CLAMP env variable should be also removed if accuracy validation show no issues

@maxnick maxnick removed this pull request from the merge queue due to a manual request Nov 5, 2025
@maxnick maxnick enabled auto-merge November 5, 2025 10:10
@isanghao
Copy link
Contributor

isanghao commented Nov 5, 2025

@v-Golubev did you validate this PR on existing model/validation set?

@v-Golubev v-Golubev disabled auto-merge November 5, 2025 11:03
@v-Golubev
Copy link
Contributor Author

v-Golubev commented Nov 5, 2025

The PR can be merged after we get accuracy validation results on real models

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@isanghao isanghao isanghao left review comments

@aobolensk aobolensk aobolensk approved these changes

@mryzhov mryzhov Awaiting requested review from mryzhov mryzhov is a code owner automatically assigned from openvinotoolkit/openvino-ie-transformations-maintainers

Assignees

@isanghao isanghao

@aobolensk aobolensk

Labels

category: GPU OpenVINO GPU plugin category: LP transformations OpenVINO Low Precision transformations category: transformations OpenVINO Runtime library - Transformations Code Freeze do_not_merge

Projects

None yet

Milestone

2025.4

Development

Successfully merging this pull request may close these issues.

5 participants

@v-Golubev @isanghao @aobolensk @moslex @maxnick