fix openai v1 (#1178)

Co-authored-by: shihaobai <42648726+shihaobai@users.noreply.github.com>

Author: sufubao

lightllm/models/qwen2/model.py

@@ -18,7 +18,7 @@ def __init__(self, kvargs):
 
     def _init_config(self):
         super()._init_config()
-        if self.config["sliding_window"] is None:
+        if self.config.get("sliding_window", None) is None:
             self.config["sliding_window"] = self.max_total_token_num
         # rename key [SYM: to be confirmed]
         return

lightllm/server/api_models.py

@@ -24,9 +24,10 @@ class Message(BaseModel):
 class Function(BaseModel):
     """Function descriptions."""
 
-    description: Optional[str] = Field(default=None, examples=[None])
     name: Optional[str] = None
-    parameters: Optional[object] = None
+    description: Optional[str] = Field(default=None, examples=[None])
+    parameters: Optional[dict] = None
+    response: Optional[dict] = None
 
 
 class Tool(BaseModel):

lightllm/server/api_openai.py

@@ -81,7 +81,7 @@ def _process_tool_call_id(
         # SGLang sets call_item.tool_index to the *local* position inside that message.
         # Therefore, the index must be corrected by using
         # `history_tool_calls_cnt + call_item.tool_index` to ensure globally unique and properly ordered.
-        tool_call_id = f"functions.{call_item.name}:{history_tool_calls_cnt+call_item.tool_index}"
+        tool_call_id = f"functions.{call_item.name}:{history_tool_calls_cnt + call_item.tool_index}"
         logger.debug(
             f"Process tool call idx, parser: {tool_call_parser}, \
             tool_call_id: {tool_call_id}, \

lightllm/server/core/objs/sampling_params.py

@@ -7,7 +7,7 @@
 
 _SAMPLING_EPS = 1e-5
 DEFAULT_INPUT_PENALTY = os.getenv("INPUT_PENALTY", "False").upper() in ["ON", "TRUE", "1"]
-SKIP_SPECIAL_TOKENS = os.getenv("SKIP_SPECIAL_TOKENS", "True").upper() in ["ON", "TRUE", "1"]
+SKIP_SPECIAL_TOKENS = os.getenv("SKIP_SPECIAL_TOKENS", "False").upper() in ["ON", "TRUE", "1"]
 
 # 从环境变量获取最大长度限制
 STOP_SEQUENCE_MAX_LENGTH = int(os.getenv("LIGHTLLM_STOP_SEQUENCE_MAX_LENGTH", 256))