Merge pull request #1057 from Todysheep/dev

feat: 更新LLMRequest类以支持自定义参数，更新payload键值添加逻辑，兼容不支持某些键值的api

Author: SengokuCola

src/llm_models/utils_model.py

@@ -124,14 +124,21 @@ def __init__(self, model: dict, **kwargs):
         self.model_name: str = model["name"]
         self.params = kwargs
 
-        self.enable_thinking = model.get("enable_thinking", False)
+        self.enable_thinking = model.get("enable_thinking", None)
         self.temp = model.get("temp", 0.7)
-        self.thinking_budget = model.get("thinking_budget", 4096)
+        self.thinking_budget = model.get("thinking_budget", None)
         self.stream = model.get("stream", False)
         self.pri_in = model.get("pri_in", 0)
         self.pri_out = model.get("pri_out", 0)
         self.max_tokens = model.get("max_tokens", global_config.model.model_max_output_length)
         # print(f"max_tokens: {self.max_tokens}")
+        custom_params_str = model.get("custom_params", "{}")
+        try:
+            self.custom_params = json.loads(custom_params_str)
+        except json.JSONDecodeError as e:
+            logger.error(f"Invalid JSON in custom_params for model '{self.model_name}': {custom_params_str}")
+            self.custom_params = {}
+
 
         # 获取数据库实例
         self._init_database()
@@ -249,28 +256,6 @@ async def _prepare_request(
         elif payload is None:
             payload = await self._build_payload(prompt)
 
-        if stream_mode:
-            payload["stream"] = stream_mode
-
-        if self.temp != 0.7:
-            payload["temperature"] = self.temp
-
-        # 添加enable_thinking参数（如果不是默认值False）
-        if not self.enable_thinking:
-            payload["enable_thinking"] = False
-
-        if self.thinking_budget != 4096:
-            payload["thinking_budget"] = self.thinking_budget
-
-        if self.max_tokens:
-            payload["max_tokens"] = self.max_tokens
-
-        # if "max_tokens" not in payload and "max_completion_tokens" not in payload:
-        # payload["max_tokens"] = global_config.model.model_max_output_length
-        # 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
-        if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
-            payload["max_completion_tokens"] = payload.pop("max_tokens")
-
         return {
             "policy": policy,
             "payload": payload,
@@ -670,18 +655,16 @@ async def _build_payload(self, prompt: str, image_base64: str = None, image_form
         if self.temp != 0.7:
             payload["temperature"] = self.temp
 
-        # 添加enable_thinking参数（如果不是默认值False）
-        if not self.enable_thinking:
-            payload["enable_thinking"] = False
+        # 仅当配置文件中存在参数时，添加对应参数
+        if self.enable_thinking is not None:
+            payload["enable_thinking"] = self.enable_thinking
 
-        if self.thinking_budget != 4096:
+        if self.thinking_budget is not None:
             payload["thinking_budget"] = self.thinking_budget
 
         if self.max_tokens:
             payload["max_tokens"] = self.max_tokens
 
-        # if "max_tokens" not in payload and "max_completion_tokens" not in payload:
-        # payload["max_tokens"] = global_config.model.model_max_output_length
         # 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
         if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
             payload["max_completion_tokens"] = payload.pop("max_tokens")