Skip to content

Cherrypick fix int8 cosine normalization for GPU CAGRA to 2.6#1724

Merged
sre-ci-robot merged 1 commit into
zilliztech:2.6from
marcelo-cjl:cherrypick-2.6-cagra-int8-cosine
Jul 17, 2026
Merged

Cherrypick fix int8 cosine normalization for GPU CAGRA to 2.6#1724
sre-ci-robot merged 1 commit into
zilliztech:2.6from
marcelo-cjl:cherrypick-2.6-cagra-int8-cosine

Conversation

@marcelo-cjl

@marcelo-cjl marcelo-cjl commented Jul 17, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Cherrypick fix: handle int8 cosine normalization for GPU CAGRA #1720 to 2.6.
  • Normalize INT8 COSINE rows with a float norm and a scaled int8 representation before GPU CAGRA build/search.
  • Rescale valid INT8 COSINE GPU CAGRA scores by 127^2 so returned COSINE scores stay in [-1, 1].
  • Add GPU_CUVS_CAGRA INT8 COSINE regression coverage.

pr: #1720
issue: #1721

Test plan

  • git diff --check -- src/common/cuvs/integration/cuvs_knowhere_index.cuh tests/ut/test_gpu_search.cc
  • clang-format-15 --dry-run --Werror src/common/cuvs/integration/cuvs_knowhere_index.cuh tests/ut/test_gpu_search.cc
  • cmake --build build/Release --target knowhere_tests -- -j16
  • knowhere_tests "Test All GPU Index" -c "Test Gpu Cagra Int8 Cosine Score Range" -s

Signed-off-by: marcelo-cjl <marcelo.chen@zilliz.com>
(cherry picked from commit e236277)
Signed-off-by: marcelo-cjl <marcelo.chen@zilliz.com>
@sre-ci-robot

Copy link
Copy Markdown
Collaborator

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: marcelo-cjl

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details Needs approval from an approver in each of these files:

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

@alexanderguzhva

Copy link
Copy Markdown
Collaborator

/lgtm

@sre-ci-robot
sre-ci-robot merged commit cd78828 into zilliztech:2.6 Jul 17, 2026
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants