Skip to content

[MM] Support mm_device_do_normalize for gemma3. - #64487

Merged
noooop merged 19 commits into
vllm-project:mainfrom
noooop:device-normalization-gemma
Nov 3, 2026
Merged

noooop merged 19 commits into
vllm-project:mainfrom
noooop:device-normalization-gemma

Conversation

@noooop

@noooop noooop commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Overview

Support mm_device_do_normalize for gemma3.

  • Make FusedMMInputNorm support the bchw Layout.

Claims

Validation

Test for FusedMMInputNorm

  • tests/model_executor/layers/test_mm_input_norm.py

Test for Models

  • tests/models/multimodal/generation_ppl_test/gemma.py

Details

The BCHW layout can be converted into the BCL layout through a simple view, so fused_mm_input_norm_triton can support this format without adding a new kernel—lucky.


Pull Request Checklist
  • I used vLLM's /pr-checklist skill. (Mandatory for agents, optional for humans).

  • AI assistance was used during the creation of this PR.

  • Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.

  • Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.

  • Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.

  • Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@mergify mergify Bot added multi-modality Related to multi-modality (#4194) qwen Related to Qwen models glm labels Oct 29, 2026
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@noooop noooop changed the title [MM] Support mm_device_do_normalize for gemma family. [MM] Support mm_device_do_normalize for gemma3/paligemma. Oct 30, 2026
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@noooop noooop changed the title [MM] Support mm_device_do_normalize for gemma3/paligemma. [MM] Support mm_device_do_normalize for gemma3. Nov 2, 2026
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@mergify

mergify Bot commented Nov 2, 2026

Copy link
Copy Markdown
Contributor

Documentation preview: https://vllm--64487.org.readthedocs.build/en/64487/

@mergify mergify Bot added the documentation Improvements or additions to documentation label Nov 2, 2026
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@noooop
noooop marked this pull request as ready for review November 2, 2026 06:19
@noooop
noooop requested a review from DarkLight1337 as a code owner November 2, 2026 06:19

@claude claude Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

@noooop
noooop requested a review from Isotr0py November 2, 2026 06:19
Comment thread vllm/model_executor/models/gemma3_mm.py Resolved
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
@DarkLight1337

Copy link
Copy Markdown
Member

/ci run

@github-actions

github-actions Bot commented Nov 2, 2026

Copy link
Copy Markdown

✅ Triggered Buildkite CI #99110 for commit d68bf009b789.

ℹ️ This PR is 2 commits behind upstream main. Merged upstream main into this branch as d68bf009b789 before starting CI. Pull the branch before pushing more changes.

@noooop

noooop commented Nov 3, 2026

Copy link
Copy Markdown
Contributor Author

/ci run

@github-actions

github-actions Bot commented Nov 3, 2026

Copy link
Copy Markdown

✅ Triggered Buildkite CI #99408 for commit ab02625ba578.

@noooop

noooop commented Nov 3, 2026

Copy link
Copy Markdown
Contributor Author

/ci run

@noooop
noooop enabled auto-merge (squash) November 3, 2026 02:56
@github-actions

github-actions Bot commented Nov 3, 2026

Copy link
Copy Markdown

✅ Triggered Buildkite CI #99426 for commit 9f4901343efa.

ℹ️ This PR is 1 commit behind upstream main. Merged upstream main into this branch as 9f4901343efa before starting CI. Pull the branch before pushing more changes.

@github-actions github-actions Bot added the ready ONLY add when PR is ready to merge/full CI is needed label Nov 3, 2026
@noooop

noooop commented Nov 3, 2026

Copy link
Copy Markdown
Contributor Author

/ci retry

@github-actions

github-actions Bot commented Nov 3, 2026

Copy link
Copy Markdown

✅ Triggered Buildkite CI #99471 for commit 86c17b5aaee2, running 1 failed step(s) from Buildkite CI #99426.

@noooop
noooop merged commit 1eec157 into vllm-project:main Nov 3, 2026
15 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation glm multi-modality Related to multi-modality (#4194) qwen Related to Qwen models ready ONLY add when PR is ready to merge/full CI is needed

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants