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 15, 11:00 UTC
[Bugfix] `use_fast` failing to be propagated to Qwen2-VL image processor (#17838)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
May 8, 2025 at 20:39 UTC
015815fe01412a26e72e6102d3358e157015cf65
3 files changed
+15
-9
vllm/model_executor/models/qwen2_5_omni_thinker.py
+5
-3
@@ -145,9 +145,11 @@ class Qwen2_5OmniThinkerProcessingInfo(Qwen2AudioProcessingInfo,
145
kwargs["fps"] = fps
146
processor = self.ctx.get_hf_processor(
147
Qwen2_5OmniProcessor,
148
- image_processor=self.get_image_processor(min_pixels=min_pixels,
149
- max_pixels=max_pixels,
150
- size=size),
148
+ image_processor=self.get_image_processor(
149
+ min_pixels=min_pixels,
150
+ max_pixels=max_pixels,
151
+ size=size,
152
+ use_fast=kwargs.get("use_fast")),
153
**kwargs,
154
)
155
if not hasattr(processor, "audio_token"):
vllm/model_executor/models/qwen2_5_vl.py
+5
-3
@@ -758,9 +758,11 @@ class Qwen2_5_VLProcessingInfo(Qwen2VLProcessingInfo):
758
759
return self.ctx.get_hf_processor(
760
Qwen2_5_VLProcessor,
761
- image_processor=self.get_image_processor(min_pixels=min_pixels,
762
- max_pixels=max_pixels,
763
- size=size),
761
+ image_processor=self.get_image_processor(
762
+ min_pixels=min_pixels,
763
+ max_pixels=max_pixels,
764
+ size=size,
765
+ use_fast=kwargs.get("use_fast")),
766
**kwargs,
767
)
768
vllm/model_executor/models/qwen2_vl.py
+5
-3
@@ -759,9 +759,11 @@ class Qwen2VLProcessingInfo(BaseProcessingInfo):
759
) -> Qwen2VLProcessor:
760
return self.ctx.get_hf_processor(
761
Qwen2VLProcessor,
762
- image_processor=self.get_image_processor(min_pixels=min_pixels,
763
- max_pixels=max_pixels,
764
- size=size),
762
+ image_processor=self.get_image_processor(
763
+ min_pixels=min_pixels,
764
+ max_pixels=max_pixels,
765
+ size=size,
766
+ use_fast=kwargs.get("use_fast")),
767
**kwargs,
768
)
769