diff --git a/vllm/entrypoints/openai/chat_completion/batch_serving.py b/vllm/entrypoints/openai/chat_completion/batch_serving.py index a0fc867050..6acc568f49 100644 --- a/vllm/entrypoints/openai/chat_completion/batch_serving.py +++ b/vllm/entrypoints/openai/chat_completion/batch_serving.py @@ -267,6 +267,7 @@ class OpenAIServingChatBatch(OpenAIServingChat): reasoning, content, _ = parser.parse( output.text, request=request, # type: ignore[arg-type] + model_output_token_ids=output.token_ids, ) if not request.include_reasoning: reasoning = None diff --git a/vllm/entrypoints/openai/responses/context.py b/vllm/entrypoints/openai/responses/context.py index 3c9a31a141..a7cb96f949 100644 --- a/vllm/entrypoints/openai/responses/context.py +++ b/vllm/entrypoints/openai/responses/context.py @@ -346,6 +346,7 @@ class ParsableContext(ConversationContext): completion.text, self.request, enable_auto_tools=self.enable_auto_tools, + model_output_token_ids=completion.token_ids, ) self.response_messages.extend( build_response_output_items( diff --git a/vllm/entrypoints/openai/responses/serving.py b/vllm/entrypoints/openai/responses/serving.py index f2e1f8e5d8..6746434a04 100644 --- a/vllm/entrypoints/openai/responses/serving.py +++ b/vllm/entrypoints/openai/responses/serving.py @@ -1063,6 +1063,7 @@ class OpenAIServingResponses(OpenAIServing): final_output.text, request, enable_auto_tools=self.enable_auto_tools, + model_output_token_ids=final_output.token_ids, ) return build_response_output_items( reasoning=reasoning,