Add Qwen3-omni offline UT #216
Conversation
Signed-off-by: Rein Yang <[email protected]>
|
The account who enabled Codex for this repo no longer has access to Codex. Please contact the admins of this repo to enable Codex again. |
|
@ywang96 Hi Roger, I’m running pytest on an H200 with a tweaked qwen3‑omni config. I’m worried it may not work on A100 or other GPUs. Could you let me know which device types are available on Buildkite? So I can update the CI yaml to test again. Thanks! |
| - pytest -s -v tests/multi_stages/qwen3_omni/ | ||
| agents: | ||
| queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU | ||
| queue: "gpu_4_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU |
There was a problem hiding this comment.
How many GB of VRAM does each GPU have, and where can I find these resource details?
There was a problem hiding this comment.
L4 has 24GB VRAM, but there is usually only about 22GB available on CI machine actually.
|
Maybe we could reuse the |
|
|
||
|
|
||
| @pytest.mark.parametrize("model", models) | ||
| def test_image_to_audio(omni_runner: type[OmniRunner], model: str) -> None: |
There was a problem hiding this comment.
I think that we could keep only one test that outputs audio to save resources.
There was a problem hiding this comment.
Should I add at least one audio test and one video/image test? The visual has more handling than audio. And qwen3-omni has some bug now don't support mixed modality.
| timeout_in_minutes: 20 | ||
| depends_on: image-build | ||
| commands: | ||
| - pytest -s -v tests/multi_stages/qwen3_omni/ |
There was a problem hiding this comment.
| - pytest -s -v tests/multi_stages/qwen3_omni/ | |
| - export VLLM_WORKER_MULTIPROC_METHOD=spawn | |
| - pytest -s -v tests/multi_stages/qwen3_omni/ |
Signed-off-by: Roger Wang <[email protected]>
Sure, I refer your PR, can add one single modality test to conftest.py later after your's merged. |
|
Maybe it's a good idea to use 2xH100 for qwen3-omni e2e test for now unless we run into resource issues. Feel free to take a look at #221 |
5 participants
PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.
Purpose
Add offline UT for qwen3-omni
Test Plan
pytest -sv tests/multi_stages/qwen3_omni/test_qwen3_omni.py
Test Result
Pass
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.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)