@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 5d712571af
Fork 0

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 16, 04:30 UTC

[Bugfix] Quick fix to make Pixtral-HF load correctly again after 39e227c7ae. (#11024)

Jeff Cook committed Dec 12, 2024 at 11:09 UTC 5d712571afd87f5cc1b6c6c25feaac5c706f3712
1 file changed +1 -5
vllm/model_executor/models/llava.py
+1 -5
@@ -218,12 +218,8 @@ class LlavaProcessor(BaseMultiModalProcessor):
218 image_processor = hf_processor.image_processor # type: ignore
219 hf_inputs = image_processor.preprocess(data['image'],
220 return_tensors="pt")
221 - is_pixtral = isinstance(hf_processor, PixtralProcessor)
221
223 - return MultiModalKwargs(
224 - **hf_inputs,
225 - is_pixtral=torch.tensor(is_pixtral),
226 - )
222 + return MultiModalKwargs(**hf_inputs)
223
224
225 class LlavaLikeConfig(Protocol):