Repository navigation
[ROCm] [CI] Fix diffusion test - #4574
Merged
Merged
Conversation
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
tjtanaa
requested review from
Isotr0py,
RuixiangMa,
SamitHuang,
ZJY0516,
david6666666,
princepride,
wtomin and
yenuo26
as code owners
June 19, 2026 15:30
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
hsliuustc0106
left a comment
Collaborator
There was a problem hiding this comment.
Test-only fix looks correct. The _get_cumem_allocator_class() wrapper prevents hipSetDevice errors on ROCm by deferring the CuMemAllocator import. All CI checks pass.
| return vllm_config | ||
|
|
||
|
|
||
| def _get_cumem_allocator_class() -> type: |
Collaborator
There was a problem hiding this comment.
Good: Lazy import wrapper prevents hipSetDevice symbol error on ROCm.
| pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.gpu] | ||
|
|
||
|
|
||
| def patch_cumem_allocator(mocker: MockerFixture): |
Collaborator
There was a problem hiding this comment.
Good: Helper function centralizes the mock patching pattern.
Collaborator
|
amd ci passed |
khairulkabir1661
pushed a commit
to khairulkabir1661/vllm-omni
that referenced
this pull request
Sep 25, 2026
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
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
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.
PLEASE FILL IN THE PR DESCRIPTION HERE.
Purpose
Fix Diffusion GPU Worker Test
https://buildkite.com/vllm/vllm-omni-amd-ci/builds/8941/list?sid=019edeb5-8b47-4515-810d-ef3d20b13b1c&tab=output
Test Plan
test-ready.yaml: https://buildkite.com/vllm/vllm-omni-amd-ci/builds/8942/list
The same test command is run on test-merge.yaml, so only test-ready.yaml results is sufficient.
vLLM Version:
vLLM-Omni Commit:
Test Result
BEFORE SUBMITTING: read CONTRIBUTING.md and run the precheck-pr skill with the code agent for a self-check against project conventions.
(anything written below this line will be removed by GitHub Actions)