function call in reasoning content (#1158)

Author: shihaobai

lightllm/server/api_models.py

@@ -68,6 +68,46 @@ class ResponseFormat(BaseModel):
     json_schema: Optional[JsonSchemaResponseFormat] = None
 
 
+class FunctionResponse(BaseModel):
+    """Function response."""
+
+    name: Optional[str] = None
+    arguments: Optional[str] = None
+
+
+class ToolCall(BaseModel):
+    """Tool call response."""
+
+    id: Optional[str] = None
+    index: Optional[int] = None
+    type: Literal["function"] = "function"
+    function: FunctionResponse
+
+
+class ChatCompletionMessageGenericParam(BaseModel):
+    role: Literal["system", "assistant", "tool", "function"]
+    content: Union[str, List[MessageContent], None] = Field(default=None)
+    tool_call_id: Optional[str] = None
+    name: Optional[str] = None
+    reasoning_content: Optional[str] = None
+    tool_calls: Optional[List[ToolCall]] = Field(default=None, examples=[None])
+
+    @field_validator("role", mode="before")
+    @classmethod
+    def _normalize_role(cls, v):
+        if isinstance(v, str):
+            v_lower = v.lower()
+            if v_lower not in {"system", "assistant", "tool", "function"}:
+                raise ValueError(
+                    "'role' must be one of 'system', 'assistant', 'tool', or 'function' (case-insensitive)."
+                )
+            return v_lower
+        raise ValueError("'role' must be a string")
+
+
+ChatCompletionMessageParam = Union[ChatCompletionMessageGenericParam, Message]
+
+
 class CompletionRequest(BaseModel):
     model: str
     # prompt: string or tokens
@@ -137,7 +177,7 @@ def apply_loaded_defaults(cls, data: Any):
 
 class ChatCompletionRequest(BaseModel):
     model: str
-    messages: List[Message]
+    messages: List[ChatCompletionMessageParam]
     function_call: Optional[str] = "none"
     temperature: Optional[float] = 1
     top_p: Optional[float] = 1.0
@@ -212,46 +252,6 @@ def apply_loaded_defaults(cls, data: Any):
         return data
 
 
-class FunctionResponse(BaseModel):
-    """Function response."""
-
-    name: Optional[str] = None
-    arguments: Optional[str] = None
-
-
-class ToolCall(BaseModel):
-    """Tool call response."""
-
-    id: Optional[str] = None
-    index: Optional[int] = None
-    type: Literal["function"] = "function"
-    function: FunctionResponse
-
-
-class ChatCompletionMessageGenericParam(BaseModel):
-    role: Literal["system", "assistant", "tool", "function"]
-    content: Union[str, List[MessageContent], None] = Field(default=None)
-    tool_call_id: Optional[str] = None
-    name: Optional[str] = None
-    reasoning_content: Optional[str] = None
-    tool_calls: Optional[List[ToolCall]] = Field(default=None, examples=[None])
-
-    @field_validator("role", mode="before")
-    @classmethod
-    def _normalize_role(cls, v):
-        if isinstance(v, str):
-            v_lower = v.lower()
-            if v_lower not in {"system", "assistant", "tool", "function"}:
-                raise ValueError(
-                    "'role' must be one of 'system', 'assistant', 'tool', or 'function' (case-insensitive)."
-                )
-            return v_lower
-        raise ValueError("'role' must be a string")
-
-
-ChatCompletionMessageParam = Union[ChatCompletionMessageGenericParam, Message]
-
-
 class UsageInfo(BaseModel):
     prompt_tokens: int = 0
     completion_tokens: Optional[int] = 0

lightllm/server/api_openai.py

@@ -260,6 +260,7 @@ async def chat_completions_impl(request: ChatCompletionRequest, raw_request: Req
 
             finish_reason = finish_reason_dict[sub_req_id]
             text = "".join(final_output_dict[sub_req_id])
+            full_text = text
 
             # Handle reasoning content
             reasoning_text = None
@@ -284,14 +285,14 @@ async def chat_completions_impl(request: ChatCompletionRequest, raw_request: Req
             tool_calls = None
             tool_choice = request.tool_choice
             tools = request.tools
-            if tool_choice != "none" and any([i in text for i in TOOLS_TAG_LIST]):
+            if tool_choice != "none" and any([i in full_text for i in TOOLS_TAG_LIST]):
                 if finish_reason == "stop":
                     finish_reason = "tool_calls"
                 try:
                     # 为 tool_call_parser 提供默认值
                     tool_parser = getattr(g_objs.args, "tool_call_parser", None) or "llama3"
                     parser = FunctionCallParser(tools, tool_parser)
-                    full_normal_text, call_info_list = parser.parse_non_stream(text)
+                    full_normal_text, call_info_list = parser.parse_non_stream(full_text)
                     tool_calls = []
                     history_tool_calls_cnt = _get_history_tool_calls_cnt(request)
                     for call_info in call_info_list: