[Bugfix] Reject tokenizer-less Qwen VL processor init (#54886)

Signed-off-by: Luyi Xiao <81631911+luyixiao95@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

tests/model_executor/test_qwen3_omni.py

@@ -144,6 +144,7 @@ def test_qwen3_omni_get_updates_use_audio_in_video(
 
     # Create a mock context
     mock_ctx = Mock(spec=InputProcessingContext)
+    mock_ctx.tokenizer = mock_tokenizer
 
     # Create processing info
     info = Qwen3OmniMoeThinkerProcessingInfo(mock_ctx)

vllm/model_executor/models/terratorch.py

@@ -169,6 +169,9 @@ class TerratorchInputBuilder(BaseDummyInputsBuilder[TerratorchProcessingInfo]):
 
 
 class TerratorchMultiModalProcessor(BaseMultiModalProcessor[TerratorchProcessingInfo]):
+    # Terratorch models, including Prithvi, consume tokenizer-free inputs.
+    requires_tokenizer = False
+
     def _get_mm_fields_config(
         self,
         hf_inputs: BatchFeature,

vllm/multimodal/processing/processor.py

@@ -8,6 +8,7 @@ from enum import Enum
 from functools import lru_cache
 from typing import (
     TYPE_CHECKING,
+    ClassVar,
     Generic,
     NamedTuple,
     Protocol,
@@ -1033,6 +1034,8 @@ class BaseMultiModalProcessor(ABC, Generic[_I]):
     Not to be confused with `transformers.ProcessorMixin`.
     """
 
+    requires_tokenizer: ClassVar[bool] = True
+
     def __init__(
         self,
         info: _I,
@@ -1042,6 +1045,13 @@ class BaseMultiModalProcessor(ABC, Generic[_I]):
     ) -> None:
         super().__init__()
 
+        if self.requires_tokenizer and info.ctx.tokenizer is None:
+            raise ValueError(
+                f"{type(self).__name__} requires a tokenizer and cannot be "
+                "initialized when `skip_tokenizer_init=True`. Disable "
+                "`--skip-tokenizer-init`."
+            )
+
         self.info = info
         self.dummy_inputs = dummy_inputs
         self.cache = cache