[Chore] Enable passing tokenizer=None into MM processor (#29724)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/glm4_1v.py

@@ -866,12 +866,6 @@ class Glm4vVisionTransformer(nn.Module):
 
 
 class Glm4vProcessingInfo(BaseProcessingInfo):
-    def get_hf_config(self):
-        return self.ctx.get_hf_config()
-
-    def get_tokenizer(self):
-        return self.ctx.tokenizer
-
     def get_supported_mm_limits(self) -> Mapping[str, int | None]:
         return {"image": None, "video": 1}
 

vllm/model_executor/models/qwen3_vl.py

@@ -615,9 +615,6 @@ class Qwen3VLProcessingInfo(Qwen2VLProcessingInfo):
             **kwargs,
         )
 
-    def get_tokenizer(self):
-        return self.ctx.tokenizer
-
     def get_image_processor(self, **kwargs: object) -> Qwen2VLImageProcessorFast:
         return self.get_hf_processor(**kwargs).image_processor
 

vllm/model_executor/models/qwen_vl.py

@@ -555,7 +555,7 @@ class QwenVLProcessor:
 
 class QwenVLProcessingInfo(BaseProcessingInfo):
     def get_tokenizer(self) -> PreTrainedTokenizer:
-        tokenizer = self.ctx.tokenizer
+        tokenizer = self.ctx.get_tokenizer()
         assert isinstance(tokenizer, PreTrainedTokenizer)
 
         return _get_tokenizer_without_image_pad(tokenizer)