[Bugfix] Fix Qwen3-Omni crash on video with no audio track when use_audio_in_video=True (#48420)
Signed-off-by: RyanJHamby <ryanhamby22@gmail.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
@@ -113,3 +113,35 @@ def test_audio_in_video_cache_correctness(model_id: str, num_videos: int) -> Non
|
||||
f" baseline : {baseline_ids}\n"
|
||||
f" cache-hit: {second_ids}"
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model_id", MODELS)
|
||||
def test_use_audio_in_video_without_audio_track(model_id: str) -> None:
|
||||
"""
|
||||
A video with no audio track, combined with `use_audio_in_video=True`,
|
||||
must raise a clear `ValueError` naming the mismatch, rather than an
|
||||
opaque `StopIteration`-derived error from the underlying HF processor.
|
||||
"""
|
||||
ctx = build_model_context(
|
||||
model_id,
|
||||
limit_mm_per_prompt={"audio": 1, "image": 0, "video": 1},
|
||||
)
|
||||
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config, cache=None)
|
||||
video_token_id = processor.info.get_hf_config().video_token_id
|
||||
|
||||
rng = np.random.RandomState(0)
|
||||
video = random_video(rng, min_frames=8, max_frames=9, min_wh=64, max_wh=65)
|
||||
|
||||
# No "audio" key at all in this request's data: this is the "video has
|
||||
# no audio track" case. (The audio limit above is 1, not 0, so that
|
||||
# `mm_counts["audio"]` is still declared for this request; a 0 limit
|
||||
# would make vLLM omit the "audio" key from `mm_counts` entirely, which
|
||||
# trips an unrelated assertion in `_apply_hf_processor_mm_only`.)
|
||||
mm_data = {"video": [video]}
|
||||
|
||||
with pytest.raises(ValueError, match="doesn't have audio track"):
|
||||
processor(
|
||||
[video_token_id],
|
||||
mm_items=processor.info.parse_mm_data(mm_data),
|
||||
hf_processor_mm_kwargs={"use_audio_in_video": True},
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user