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, 11:15 UTC
[Multimodal][Render] Skip mm processor initialization and warmup for text-only mode (#41246)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Apr 30, 2026 at 05:16 UTC
1628239eb234739646e21c4053e3fa652058e245
1 file changed
+1
-1
vllm/renderers/base.py
+1
-1
@@ -105,7 +105,7 @@ class BaseRenderer(ABC, Generic[_T]):
105
self._process_multimodal_async = make_async(
106
self._process_multimodal, executor=self._mm_executor
107
)
108
- if config.model_config.is_multimodal_model:
108
+ if mm_registry.supports_multimodal_inputs(config.model_config):
109
mm_processor_cache = mm_registry.processor_cache_from_config(config)
110
111
# Deep-copy the tokenizer so the multimodal processor gets its