[Bugfix][VLM] fix llava processor (#15285)

Signed-off-by: Mengqing Cao <[email protected]>

Author: MengqingCao

vllm/model_executor/models/llava.py

@@ -233,7 +233,13 @@ def get_dummy_processor_inputs(
 class LlavaProcessingInfo(BaseLlavaProcessingInfo):
 
     def get_hf_processor(self, **kwargs: object):
-        return self.ctx.get_hf_processor(LlavaProcessor, **kwargs)
+        hf_processor = self.ctx.get_hf_processor(LlavaProcessor, **kwargs)
+        # In case patch_size is omitted from `processor_config.json`
+        # e.g. for E5-V: https://huggingface.co/royokong/e5-v
+        if hf_processor.patch_size is None:
+            patch_size = self.get_vision_encoder_info().get_patch_size()
+            hf_processor.patch_size = patch_size
+        return hf_processor
 
 
 class BaseLlavaMultiModalProcessor(BaseMultiModalProcessor[_I]):