A high-throughput and memory-efficient inference and serving engine for LLMs
[BugFix] Use served model name in gemma4 audio-tower error message (#44828)
Signed-off-by: Sungjae Lee <33976427+llsj14@users.noreply.github.com> Signed-off-by: Sungjae Lee <sung-jae.lee@navercorp.com>
Sungjae Lee committed
Jun 8, 2026 at 15:58 UTC
469f3dcf1d70349d0776ef82bdc5c4aff20a93f2
1 file changed
+5
-1
vllm/model_executor/models/gemma4_mm.py
+5
-1
@@ -34,6 +34,7 @@ from transformers.models.gemma4.configuration_gemma4 import (
34
)
35
36
from vllm.config import VllmConfig
37
+from vllm.config.model import get_served_model_name
38
from vllm.config.multimodal import BaseDummyOptions, VideoDummyOptions
39
from vllm.inputs import MultiModalDataDict
40
from vllm.logger import init_logger
@@ -217,7 +218,10 @@ class Gemma4ProcessingInfo(BaseProcessingInfo):
218
and num_items > 0
219
and self.get_hf_config().audio_config is None
220
):
220
- model = self.ctx.model_config.model
221
+ model_config = self.ctx.model_config
222
+ model = get_served_model_name(
223
+ model_config.model, model_config.served_model_name
224
+ )
225
raise ValueError(
226
f"Audio input was provided but the model "
227
f"'{model}' does not have an audio tower. "