[V1][Structured Output] Minor modification to `_validate_structured_output()` (#16748)

Signed-off-by: shen-shanshan <467638484@qq.com>

vllm/v1/engine/processor.py

@@ -149,6 +149,7 @@ class Processor:
             "xgrammar", "xgrammar:disable-any-whitespace", "guidance",
             "guidance:disable-any-whitespace", "auto"
         ]
+
         engine_level_backend = self.decoding_config.guided_decoding_backend
         if engine_level_backend not in supported_backends:
             raise ValueError(f"Only {supported_backends} structured output is "
@@ -169,8 +170,15 @@ class Processor:
         if engine_level_backend.startswith("xgrammar"):
             # xgrammar with no fallback
             validate_xgrammar_grammar(params)
-            params.guided_decoding.backend = engine_level_backend
-        elif engine_level_backend == "auto":
+        elif engine_level_backend.startswith("guidance"):
+            # TODO: ideally we would have the LLTokenizer here as Lark syntax
+            # allows <|special_token|> and similar, see
+            # https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md#special-tokens
+            # Without tokenizer these are disallowed in grammars.
+            validate_guidance_grammar(params, tokenizer=None)
+        else:
+            # NOTE: engine_level_backend must be "auto" here, because we have
+            # checked supported_backends above.
             # "auto" is an opt-in to opinionated behavior where we try to
             # choose a backend based on request contents. This is not the
             # default as it is less predictable and subject to change
@@ -183,14 +191,6 @@ class Processor:
                 # are not supported in xgrammar. Fall back to guidance.
                 params.guided_decoding.backend = "guidance"
 
-        if engine_level_backend.startswith("guidance"):
-            # TODO ideally we would have the LLTokenizer here as Lark syntax
-            # allows <|special_token|> and similar, see
-            # https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md#special-tokens
-            # Without tokenizer these are disallowed in grammars.
-            validate_guidance_grammar(params, tokenizer=None)
-            params.guided_decoding.backend = engine_level_backend
-
     def process_inputs(
         self,
         request_id: str,