Adapts Qwen3-Omni PD disaggregation to current config-refactor#2947
Open
spencerr221 wants to merge 16 commits into
Open
Adapts Qwen3-Omni PD disaggregation to current config-refactor#2947spencerr221 wants to merge 16 commits into
spencerr221 wants to merge 16 commits into
Conversation
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
Contributor
|
Can you add PD separation performance test cases? |
gcanlin
reviewed
Apr 20, 2026
gcanlin
reviewed
Apr 20, 2026
Collaborator
hsliuustc0106
left a comment
hsliuustc0106
left a comment
There was a problem hiding this comment.
please update the docs as well
Collaborator
hsliuustc0106
left a comment
hsliuustc0106
left a comment
There was a problem hiding this comment.
BLOCKING:
- Gate Check — pre-commit is failing. Please fix pre-commit issues before proceeding with review.
spencerr221
force-pushed
the
adapt_config
branch
from
8e7d305 to
074e448
Compare
hsliuustc0106
requested changes
Apr 21, 2026
Collaborator
hsliuustc0106
left a comment
hsliuustc0106
left a comment
There was a problem hiding this comment.
BLOCKING:
- Gate Check — pre-commit still failing (end-of-file-fixer on
test_qwen3_omni_expansion.py, E501 line too long instage_config.py:757, ruff format). Please runpre-commit run --all-fileslocally and push the fix.
Non-blocking:
- PR description Test Plan / Test Result sections are empty. Please add CI results or local run output showing the PD expansion tests pass (e.g.
test_pd_disaggregation.py,test_config_factory.py::test_merge_pipeline_deploy_with_pd_disaggregation).
spencerr221
force-pushed
the
adapt_config
branch
3 times, most recently
from
fd9136c to
cd5e8d0
Compare
Gaohan123
added
ready
spencerr221
force-pushed
the
adapt_config
branch
from
3407e76 to
701b6ff
Compare
yenuo26
reviewed
Apr 24, 2026
yenuo26
reviewed
Apr 24, 2026
1 task
spencerr221
force-pushed
the
adapt_config
branch
from
701b6ff to
4c0eb2b
Compare
spencerr221
changed the title
spencerr221
force-pushed
the
adapt_config
branch
2 times, most recently
from
3fd973b to
8a7050e
Compare
Gaohan123
reviewed
Apr 26, 2026
| if CONFIG_FILE_PATH is None: | ||
| print( | ||
| "No --test-config-file in argv, using default: tests/dfx/perf/tests/test_qwen_omni.json " | ||
| "(override with e.g. --test-config-file tests/dfx/perf/tests/test_tts.json)" |
| resources: | ||
| limits: | ||
| nvidia.com/gpu: 2 | ||
| nvidia.com/gpu: 3 |
Collaborator
There was a problem hiding this comment.
I suggest you add a separate label for PD test rather than all omni test with 3 cards
Collaborator
There was a problem hiding this comment.
I suggest you add a separate label for PD test rather than all omni test with 3 cards
if you separete job for PD test, maybe you can use job name Omni · Function Test with 2 H100 and Omni · Function Test with 3 H100
spencerr221
force-pushed
the
adapt_config
branch
2 times, most recently
from
316e921 to
fa080f7
Compare
hsliuustc0106
removed
ready
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
…th mooncake. Signed-off-by: Liubingyu <liubingyu62@gmail.com>
spencerr221
requested review from
ZeldaHuang,
congw729,
david6666666,
linyueqian,
princepride,
tzhouam and
yuanheng-zhao
as code owners
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
This PR adapts Qwen3-Omni PD (prefill-decode) separation to the new config-refactor flow introduced by #2383, without adding a new deploy YAML.
It adds
pd_separationsupport to the deploy-based config pipeline so the existingvllm_omni/deploy/qwen3_omni_moe.yamlcan dynamically expand the original 3-stage Qwen3-Omni pipeline into a 4-stage PD layout at merge time. When enabled, the thinker stage is split into prefill and decode stages, downstream stage IDs and connectors are remapped, and KV transfer settings are injected from deploy config.This keeps the new
pipeline+deploymodel intact, preserves the existing PD detection/runtime logic, and avoids introducing a separate PD-specific config file after the config refactor.Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model. Please runmkdocs serveto sync the documentation editions to./docs.BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)