[v0][structured output] Support reasoning output (#12955)

Signed-off-by: Ce Gao <cegao@tensorchord.ai>
2025-03-03 03:49:42 +08:00
parent bc6ccb9878
commit bf33700ecd
16 changed files with 400 additions and 76 deletions
--- a/vllm/engine/llm_engine.py
+++ b/vllm/engine/llm_engine.py
@@ -2048,10 +2048,15 @@ class LLMEngine:
            guided_decoding.backend = guided_decoding.backend or \
                self.decoding_config.guided_decoding_backend

+            logger.debug("Reasoning backend: %s",
+                         self.decoding_config.reasoning_backend)
+
            processor = get_local_guided_decoding_logits_processor(
                guided_params=guided_decoding,
                tokenizer=tokenizer,
-                model_config=self.model_config)
+                model_config=self.model_config,
+                reasoning_backend=self.decoding_config.reasoning_backend,
+            )
            if processor:
                logits_processors.append(processor)