[Structured Outputs][V1] Skipping with models doesn't contain tokenizers (#20365)

Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Co-authored-by: Nick Hill <nhill@redhat.com>

vllm/v1/engine/processor.py

@@ -152,6 +152,11 @@ class Processor:
         if not params.guided_decoding or not self.decoding_config:
             return
 
+        if self.model_config.skip_tokenizer_init and params.guided_decoding:
+            raise ValueError(
+                "Structured outputs requires a tokenizer so it can't be used with 'skip_tokenizer_init'"  # noqa: E501
+            )
+
         engine_level_backend = self.decoding_config.backend
         if params.guided_decoding.backend:
             # Request-level backend selection is not supported in V1.