Skip to content

[DRAFT] Added possibility to launch int8 with group_size and scale estimation - #4273

Draft
AsyaPronina wants to merge 2 commits into
openvinotoolkit:developfrom
AsyaPronina:int8_gs_and_se
Draft

AsyaPronina wants to merge 2 commits into
openvinotoolkit:developfrom
AsyaPronina:int8_gs_and_se

Conversation

@AsyaPronina

Copy link
Copy Markdown
Collaborator

Changes

Reason for changes

Related tickets

Tests

@AsyaPronina
AsyaPronina requested a review from a team as a code owner October 19, 2026 09:23
Copilot AI balanced review requested due to automatic review settings October 19, 2026 09:23
@AsyaPronina
AsyaPronina marked this pull request as draft October 19, 2026 09:23
@github-actions github-actions Bot added documentation Improvements or additions to documentation NNCF OpenVINO Pull requests that updates NNCF OpenVINO labels Oct 19, 2026
@github-actions github-actions Bot added the API Public API-impacting changes label Oct 19, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

🟡 Changes recommended

INT8 scale estimation can erase weights with zero calibration activations and overflow on signed INT8 targets.

2 open issues
What changed in this PR

Extends OpenVINO INT8 weight compression with group-wise quantization and calibration-based scale estimation.

Changes:

  • Enables INT8 grouping while retaining per-channel compression for multi-axis weights.
  • Enables INT8 scale estimation with backend-specific validation.
  • Adds OpenVINO tests and usage examples.
File Description
tests/​openvino/​native/​quantization/​test_weights_compression.py Tests grouping, convolution fallback, and scale estimation.
src/​nncf/​quantization/​quantize_model.py Validates backend support and documents grouping.
src/​nncf/​quantization/​algorithms/​weight_compression/​scale_estimation.py Allows estimation for configurable bit widths.
src/​nncf/​quantization/​algorithms/​weight_compression/​algorithm.py Configures INT8 estimation and per-channel fallback.
docs/​usage/​post_training_compression/​weights_compression/​Usage.md Documents INT8 options and examples.

Review effort: Balanced

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

🔵 Needs a closer look

Numerical accuracy and grouped-weight execution need backend runtime validation before approval.

No open issues 🎉

2 resolved since last review

Review effort: Balanced

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

API Public API-impacting changes documentation Improvements or additions to documentation NNCF OpenVINO Pull requests that updates NNCF OpenVINO

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants