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 6, 23:00 UTC
[Bugfix] Fix processor signature (#36630)
Signed-off-by: raushan <raushan@huggingface.co> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Raushan Turganbay committed
Mar 10, 2026 at 14:20 UTC
8850738b700cca34448fbafbc8ac41bcad5a2e17
2 files changed
+12
-6
vllm/transformers_utils/processors/glm4v.py
+6
-3
@@ -28,8 +28,11 @@ class GLM4VProcessor(ProcessorMixin):
28
self,
29
tokenizer: PreTrainedTokenizer,
30
image_size: int,
31
+ image_processor: GLM4VImageProcessorFast | None = None,
32
) -> None:
33
self.tokenizer = tokenizer
33
- self.image_processor = GLM4VImageProcessorFast(
34
- size={"width": image_size, "height": image_size}
35
- )
34
+ if image_processor is None:
35
+ image_processor = GLM4VImageProcessorFast(
36
+ size={"width": image_size, "height": image_size}
37
+ )
38
+ self.image_processor = image_processor
vllm/transformers_utils/processors/qwen_vl.py
+6
-3
@@ -29,11 +29,14 @@ class QwenVLProcessor(ProcessorMixin):
29
self,
30
tokenizer: QwenVLTokenizer,
31
image_size: int,
32
+ image_processor: QwenVLImageProcessorFast | None = None,
33
) -> None:
34
self.tokenizer = tokenizer
34
- self.image_processor = QwenVLImageProcessorFast(
35
- size={"width": image_size, "height": image_size}
36
- )
35
+ if image_processor is None:
36
+ image_processor = QwenVLImageProcessorFast(
37
+ size={"width": image_size, "height": image_size}
38
+ )
39
+ self.image_processor = image_processor
40
41
@property
42
def image_start_tag(self) -> str: