fix: normalize system messages before model calls

offline-backend-20260512/backend/packages/harness/deerflow/models/vllm_provider.py

@@ -32,9 +32,11 @@ from langchain_core.messages import (
 )
 from langchain_core.messages.tool import tool_call_chunk
 from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
-from langchain_openai import ChatOpenAI
-from langchain_openai.chat_models.base import _create_usage_metadata
-
+from langchain_openai import ChatOpenAI
+from langchain_openai.chat_models.base import _create_usage_metadata
+
+from deerflow.models.message_ordering import normalize_chat_payload_messages
+
 
 def _normalize_vllm_chat_template_kwargs(payload: dict[str, Any]) -> None:
     """Map DeerFlow's legacy ``thinking`` toggle to vLLM/Qwen's ``enable_thinking``.
@@ -182,13 +184,14 @@ class VllmChatModel(ChatOpenAI):
             for payload_msg, orig_msg in zip(payload_messages, original_messages):
                 if payload_msg.get("role") == "assistant" and isinstance(orig_msg, AIMessage):
                     _restore_reasoning_field(payload_msg, orig_msg)
-        else:
-            ai_messages = [message for message in original_messages if isinstance(message, AIMessage)]
-            assistant_payloads = [message for message in payload_messages if message.get("role") == "assistant"]
-            for payload_msg, ai_msg in zip(assistant_payloads, ai_messages):
-                _restore_reasoning_field(payload_msg, ai_msg)
-
-        return payload
+        else:
+            ai_messages = [message for message in original_messages if isinstance(message, AIMessage)]
+            assistant_payloads = [message for message in payload_messages if message.get("role") == "assistant"]
+            for payload_msg, ai_msg in zip(assistant_payloads, ai_messages):
+                _restore_reasoning_field(payload_msg, ai_msg)
+
+        normalize_chat_payload_messages(payload)
+        return payload
 
     def _create_chat_result(self, response: dict | openai.BaseModel, generation_info: dict | None = None) -> ChatResult:
         """Preserve vLLM reasoning on non-streaming responses."""