A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 12:15 UTC
[Bugfix] Update Ultravox compatibility (#29588)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Nov 27, 2025 at 17:36 UTC
00d3310d2d00d021d2e8f5f00e31b51d30f0413e
1 file changed
+5
vllm/model_executor/models/ultravox.py
+5
@@ -116,7 +116,12 @@ class UltravoxProcessingInfo(BaseProcessingInfo):
116
117
def get_feature_extractor(self, **kwargs: object) -> WhisperFeatureExtractor:
118
hf_processor = self.get_hf_processor(**kwargs)
119
+
120
+ # Changed in https://huggingface.co/fixie-ai/ultravox-v0_5-llama-3_2-1b/commit/9a3c571b8fdaf1e66dd3ea61bbcb6db5c70a438e
121
audio_processor = hf_processor.audio_processor # type: ignore
122
+ if isinstance(audio_processor, WhisperFeatureExtractor):
123
+ return audio_processor
124
+
125
feature_extractor = audio_processor.feature_extractor # type: ignore
126
assert isinstance(feature_extractor, WhisperFeatureExtractor)
127
return feature_extractor