fix fc stream

Author: shihaobai

lightllm/server/api_models.py

@@ -193,7 +193,7 @@ class DeltaMessage(BaseModel):
 class ChatCompletionStreamResponseChoice(BaseModel):
     index: int
     delta: DeltaMessage
-    finish_reason: Optional[Literal["stop", "length"]] = None
+    finish_reason: Optional[Literal["stop", "length", "tool_calls"]] = None
 
 
 class ChatCompletionStreamResponse(BaseModel):

lightllm/server/api_openai.py

@@ -281,7 +281,7 @@ async def stream_results() -> AsyncGenerator[bytes, None]:
             if request.tool_choice != "none" and request.tools:
                 delta = request_output
                 group_request_id = convert_sub_id_to_group_id(sub_req_id)
-                index = metadata["id"]
+                index = sub_req_id
                 finish_reason = finish_status.get_finish_reason()
 
                 if index not in parser_dict:

lightllm/server/function_call_parser.py

@@ -307,7 +307,7 @@ def parse_streaming_increment(self, new_text: str, tools: List[Tool]) -> Streami
                 if cur_arguments:
                     # Calculate how much of the arguments we've already streamed
                     sent = len(self.streamed_args_for_tool[self.current_tool_id])
-                    cur_args_json = json.dumps(cur_arguments)
+                    cur_args_json = json.dumps(cur_arguments, ensure_ascii=False)
                     prev_arguments = None
                     if self.current_tool_id < len(self.prev_tool_call_arr):
                         prev_arguments = self.prev_tool_call_arr[self.current_tool_id].get("arguments")
@@ -330,7 +330,7 @@ def parse_streaming_increment(self, new_text: str, tools: List[Tool]) -> Streami
 
                     # If the tool is still being parsed, send incremental changes
                     elif prev_arguments:
-                        prev_args_json = json.dumps(prev_arguments)
+                        prev_args_json = json.dumps(prev_arguments, ensure_ascii=False)
                         if cur_args_json != prev_args_json:
                             prefix = _find_common_prefix(prev_args_json, cur_args_json)
                             argument_diff = prefix[sent:]