Skip to content
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.

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

Returned OV 2024.3 references for OV 2024.5 in Weight Compression tests #3065

Merged
merged 2 commits into from
Nov 11, 2024

Conversation

ljaljushkin
Copy link
Contributor

@ljaljushkin ljaljushkin commented Nov 7, 2024

Changes

Revert of #2954
Use references for the OV 2024.3 when run conformance tests with OV 2024.5.

Reason for changes

According to the CPU team:

  1. In 24.4 time frame, RMS impl has been enabled ([CPU] Add RMSNorm jit implementation openvino#26147), which introduce some accuracy change due to vrsqrtss which is done by approximation.
  2. In 24.5 time frame, fix has been applied ([CPU]Fix rms accuracy issue with vrsqrt openvino#26817) to use vsqrtss instead. With that we can see the result can be covered to that prior to [CPU] Add RMSNorm jit implementation openvino#26147
    With that (in 24.5), the accuracy result is recovered to that prior to RMS impl [CPU] Add RMSNorm jit implementation openvino#26147. Let's say that 24.4 may has some accuracy issue in some case. 24.5 fix that issue.

Related tickets

156605
156776
156237
151260

Tests

  • 47 build of weekly/job/openvino-nightly/job/post_training_weight_compression
    image
  • 48 build for xfail tests
    image

@github-actions github-actions bot added the NNCF PTQ Pull requests that updates NNCF PTQ label Nov 7, 2024
@ljaljushkin ljaljushkin requested a review from alexsu52 November 7, 2024 14:33
@ljaljushkin ljaljushkin marked this pull request as ready for review November 7, 2024 14:33
@ljaljushkin ljaljushkin requested a review from a team as a code owner November 7, 2024 14:34
Copy link
Contributor

@alexsu52 alexsu52 left a comment

Choose a reason for hiding this comment

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

Please update metrics for xfail tests as well to keep history of changes.

@ljaljushkin ljaljushkin requested a review from alexsu52 November 8, 2024 12:27
@alexsu52 alexsu52 merged commit 68983a7 into openvinotoolkit:develop Nov 11, 2024
14 checks passed
nikita-savelyevv pushed a commit to nikita-savelyevv/nncf that referenced this pull request Dec 11, 2024
…ts (openvinotoolkit#3065)

### Changes

Revert of openvinotoolkit#2954
Use [references for the OV
2024.3](https://github.com/ljaljushkin/nncf_pytorch/blob/873fbb80ab07bd41a3b1a2e21f68bc1a7a394810/tests/post_training/data/wc_reference_data.yaml)
when run conformance tests with OV 2024.5.

### Reason for changes

According to the CPU team: 
> 1) In 24.4 time frame, RMS impl has been enabled
(openvinotoolkit/openvino#26147), which
introduce some accuracy change due to vrsqrtss which is done by
approximation.
> 2) In 24.5 time frame, fix has been applied
(openvinotoolkit/openvino#26817) to use vsqrtss
instead. With that we can see the result can be covered to that prior to
openvinotoolkit/openvino#26147
> With that (in 24.5), the accuracy result is recovered to that prior to
RMS impl openvinotoolkit/openvino#26147. Let's
say that 24.4 may has some accuracy issue in some case. 24.5 fix that
issue.

### Related tickets

156605
156776
156237
151260

### Tests

- [x] 47 build of
weekly/job/openvino-nightly/job/post_training_weight_compression

![image](https://github.com/user-attachments/assets/10271e74-4e0b-4506-917a-306481b500e3)
- [x] 48 build for xfail tests

![image](https://github.com/user-attachments/assets/dc5a9424-35fa-4015-965a-ca08fa46e4b7)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Code Freeze NNCF PTQ Pull requests that updates NNCF PTQ
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants