xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 03:05:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix Qwen Omni audio inference (#27920)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung 2025-11-02 13:06:05 +08:00 committed by GitHub
parent 758ea2e980
commit 853a8eb53b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 2 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/models/qwen2_5_omni_thinker.py
View File

@ -130,6 +130,8 @@ class Qwen2_5OmniAudioFeatureInputs(TensorSchema):
TensorShape("nmb", "tsl", dynamic_dims={"tsl"}),
]
audio_feature_lengths: Annotated[torch.Tensor, TensorShape("na")]
feature_attention_mask: Annotated[
torch.Tensor | list[torch.Tensor],
TensorShape("na", "msl", dynamic_dims={"msl"}),
@ -732,13 +734,6 @@ class Qwen2_5OmniConditionalGenerationMixin:
input_features = audio_input["input_features"]
audio_feature_lengths = audio_input["audio_feature_lengths"]
if audio_feature_lengths.shape[0] == 1:
audio_feature_lengths = audio_feature_lengths.squeeze(0)
elif audio_feature_lengths.shape[1] == 1:
audio_feature_lengths = audio_feature_lengths.squeeze(1)
else:
raise AssertionError(audio_feature_lengths.shape)
audio_feat_lengths, audio_output_lengths = (
self.audio_tower._get_feat_extract_output_lengths(audio_feature_lengths)
)

3
vllm/model_executor/models/qwen3_omni_moe_thinker.py
View File

@ -99,7 +99,6 @@ from .utils import (
AutoWeightsLoader,
WeightsMapper,
_merge_multimodal_embeddings,
flatten_bn,
maybe_prefix,
)
from .vision import (
@ -1065,8 +1064,6 @@ class Qwen3OmniMoeConditionalGenerationMixin(Qwen2_5OmniConditionalGenerationMix
input_features = audio_input["input_features"]
audio_feature_lengths = audio_input["audio_feature_lengths"]
audio_feature_lengths = flatten_bn(audio_feature_lengths, concat=True)
audio_feat_lengths, audio_output_lengths = _get_feat_extract_output_lengths(
audio_feature_lengths
)