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 4, 15:00 UTC
[Bugfix] use `served_model_name` for multimodal error message (#41003)
Signed-off-by: Moritz Sanft <58110325+msanft@users.noreply.github.com>
Moritz Sanft committed
Apr 27, 2026 at 17:22 UTC
2c06cf3486a67efcbdf265b8a183f9ed836cebb7
2 files changed
+18
-1
tests/multimodal/test_registry.py
+16
@@ -5,6 +5,8 @@ Unit tests for MultiModalRegistry.supports_multimodal_inputs and
5
Qwen2.5-VL visual component loading behavior.
6
"""
7
8
+from types import SimpleNamespace
9
+
10
import pytest
11
12
from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -32,3 +34,17 @@ def test_supports_multimodal_inputs(model_id, limit_mm_per_prompt, expected):
34
limit_mm_per_prompt=limit_mm_per_prompt,
35
)
36
assert MULTIMODAL_REGISTRY.supports_multimodal_inputs(ctx.model_config) is expected
37
+
38
+
39
+def test_create_processor_error_uses_served_model_name():
40
+ model_config = SimpleNamespace(
41
+ is_multimodal_model=False,
42
+ model="/path/to/model/weights",
43
+ served_model_name="friendly-model-name",
44
+ )
45
+
46
+ with pytest.raises(
47
+ ValueError,
48
+ match="friendly-model-name is not a multimodal model",
49
+ ):
50
+ MULTIMODAL_REGISTRY.create_processor(model_config)
vllm/multimodal/registry.py
+2
-1
@@ -207,7 +207,8 @@ class MultiModalRegistry:
207
Create a multi-modal processor for a specific model and tokenizer.
208
"""
209
if not model_config.is_multimodal_model:
210
- raise ValueError(f"{model_config.model} is not a multimodal model")
210
+ model_name = model_config.served_model_name or model_config.model
211
+ raise ValueError(f"{model_name} is not a multimodal model")
212
213
model_cls = self._get_model_cls(model_config)
214
factories = model_cls._processor_factory