feat[resources]: 支持chat_completion调用ERNIE-Functions-8K

python/qianfan/model/model.py

@@ -535,22 +535,18 @@ def get_res(self) -> Union[ChatCompletion, Completion, Embedding, Text2Image]:
             log_warn("service status unknown, service could be unavailable.")
         if self.service_type == ServiceType.Chat:
             return ChatCompletion(
-                model=(self.model.name if self.model is not None else None),
                 endpoint=self.endpoint,
             )
         elif self.service_type == ServiceType.Completion:
             return Completion(
-                model=(self.model.name if self.model is not None else None),
                 endpoint=self.endpoint,
             )
         elif self.service_type == ServiceType.Embedding:
             return Embedding(
-                model=(self.model.name if self.model is not None else None),
                 endpoint=self.endpoint,
             )
         elif self.service_type == ServiceType.Text2Image:
             return Text2Image(
-                model=(self.model.name if self.model is not None else None),
                 endpoint=self.endpoint,
             )
         else:

python/qianfan/resources/llm/base.py

@@ -163,11 +163,11 @@ def __init__(
         self, version: Optional[Literal["1", "2", 1, 2]] = None, **kwargs: Any
     ) -> None:
         self._version = str(version) if version else "1"
-        self._real = self._real_base(self._version)(**kwargs)
-        self._backup = self._real_base("1")(**kwargs)
+        self._real = self._real_base(self._version, **kwargs)(**kwargs)
+        self._backup = self._real_base("1", **kwargs)(**kwargs)
 
     @classmethod
-    def _real_base(cls, version: str) -> Type[BaseResource]:
+    def _real_base(cls, version: str, **kwargs: Any) -> Type[BaseResource]:
         """
         return the real base class
         """

python/qianfan/resources/llm/chat_completion.py

@@ -34,6 +34,7 @@
     BatchRequestFuture,
     VersionBase,
 )
+from qianfan.resources.llm.function import Function
 from qianfan.resources.tools.tokenizer import Tokenizer
 from qianfan.resources.typing import JsonBody, QfLLMInfo, QfMessages, QfResponse, QfRole
 from qianfan.utils.logging import log_error, log_info
@@ -1537,7 +1538,24 @@ class ChatCompletion(VersionBase):
     _real: Union[_ChatCompletionV1, _ChatCompletionV2]
 
     @classmethod
-    def _real_base(cls, version: str) -> Type:
+    def _real_base(cls, version: str, **kwargs: Any) -> Type:
+        # convert to qianfan.Function
+        if kwargs.get("use_function"):
+            return Function
+        else:
+            model = kwargs.get("model") or ""
+            func_model_info_list = {
+                k.lower(): v for k, v in Function._supported_models().items()
+            }
+            func_model_info = func_model_info_list.get(model.lower())
+            if model and func_model_info:
+                if func_model_info and func_model_info.endpoint:
+                    return Function
+            endpoint = kwargs.get("endpoint", "")
+            for m in func_model_info_list.values():
+                if endpoint and m.endpoint == endpoint:
+                    return Function
+
         if version == "1":
             return _ChatCompletionV1
         elif version == "2":

python/qianfan/resources/llm/function.py

@@ -80,6 +80,8 @@ def _supported_models(cls) -> Dict[str, QfLLMInfo]:
                     "user_id",
                     "stop",
                     "max_output_tokens",
+                    "enable_user_memory",
+                    "user_memory_extract_level",
                 },
                 max_input_chars=11200,
                 max_input_tokens=7168,