isa-model 0.2.8__py3-none-any.whl → 0.3.0__py3-none-any.whl

isa_model/inference/services/llm/openai_llm_service.py

@@ -65,17 +65,21 @@ class OpenAILLMService(BaseLLMService):
         
         return bound_service
     
-    async def ainvoke(self, prompt: Union[str, List[Any], Any]) -> str:
+    async def ainvoke(self, prompt: Union[str, List[Any], Any]) -> Union[str, Any]:
         """Universal invocation method"""
         if isinstance(prompt, str):
             return await self.acompletion(prompt)
         elif isinstance(prompt, list):
+            if not prompt:
+                raise ValueError("Empty message list provided")
+            
             # 检查是否是 LangGraph 消息对象
-            if prompt and hasattr(prompt[0], 'content'):
+            first_msg = prompt[0]
+            if hasattr(first_msg, 'content') and hasattr(first_msg, 'type'):
                 # 转换 LangGraph 消息对象为标准格式
                 converted_messages = []
                 for msg in prompt:
-                    if hasattr(msg, 'type'):
+                    if hasattr(msg, 'type') and hasattr(msg, 'content'):
                         # LangGraph 消息对象
                         msg_dict = {"content": msg.content}
                         
@@ -106,14 +110,39 @@ class OpenAILLMService(BaseLLMService):
                             msg_dict["role"] = "user"  # 默认为用户消息
                         
                         converted_messages.append(msg_dict)
-                    else:
+                    elif isinstance(msg, dict):
                         # 已经是字典格式
                         converted_messages.append(msg)
+                    else:
+                        # 处理其他类型（如字符串）
+                        converted_messages.append({"role": "user", "content": str(msg)})
                 
-                return await self.achat(converted_messages)
-            else:
+                # 如果绑定了工具，返回 AIMessage 对象以兼容 LangGraph
+                if self._has_bound_tools():
+                    return await self.achat_with_message_response(converted_messages)
+                else:
+                    return await self.achat(converted_messages)
+            elif isinstance(first_msg, dict):
                 # 标准字典格式的消息
-                return await self.achat(prompt)
+                if self._has_bound_tools():
+                    return await self.achat_with_message_response(prompt)
+                else:
+                    return await self.achat(prompt)
+            else:
+                # 处理其他格式，如字符串列表
+                converted_messages = []
+                for msg in prompt:
+                    if isinstance(msg, str):
+                        converted_messages.append({"role": "user", "content": msg})
+                    elif isinstance(msg, dict):
+                        converted_messages.append(msg)
+                    else:
+                        converted_messages.append({"role": "user", "content": str(msg)})
+                
+                if self._has_bound_tools():
+                    return await self.achat_with_message_response(converted_messages)
+                else:
+                    return await self.achat(converted_messages)
         else:
             raise ValueError("Prompt must be a string or a list of messages")
     
@@ -161,6 +190,86 @@ class OpenAILLMService(BaseLLMService):
             logger.error(f"Error in chat completion: {e}")
             raise
     
+    async def achat_with_message_response(self, messages: List[Dict[str, str]]) -> Any:
+        """Chat completion method that returns message object for LangGraph compatibility"""
+        try:
+            temperature = self.config.get("temperature", 0.7)
+            max_tokens = self.config.get("max_tokens", 1024)
+            
+            kwargs = {
+                "model": self.model_name,
+                "messages": messages,
+                "temperature": temperature,
+                "max_tokens": max_tokens
+            }
+            
+            # Add tools if bound
+            if self._has_bound_tools():
+                kwargs["tools"] = self._get_bound_tools()
+                kwargs["tool_choice"] = "auto"
+            
+            response = await self.client.chat.completions.create(**kwargs)
+            
+            if response.usage:
+                self.last_token_usage = {
+                    "prompt_tokens": response.usage.prompt_tokens,
+                    "completion_tokens": response.usage.completion_tokens,
+                    "total_tokens": response.usage.total_tokens
+                }
+                
+                # Update total usage
+                self.total_token_usage["prompt_tokens"] += self.last_token_usage["prompt_tokens"]
+                self.total_token_usage["completion_tokens"] += self.last_token_usage["completion_tokens"]
+                self.total_token_usage["total_tokens"] += self.last_token_usage["total_tokens"]
+                self.total_token_usage["requests_count"] += 1
+            
+            # Create a LangGraph-compatible message object
+            message = response.choices[0].message
+            
+            # Try to import LangGraph message classes
+            try:
+                from langchain_core.messages import AIMessage
+                
+                # Create AIMessage with tool calls if present
+                if message.tool_calls:
+                    tool_calls = []
+                    for tc in message.tool_calls:
+                        tool_calls.append({
+                            "name": tc.function.name,
+                            "args": json.loads(tc.function.arguments),
+                            "id": tc.id
+                        })
+                    
+                    return AIMessage(
+                        content=message.content or "",
+                        tool_calls=tool_calls
+                    )
+                else:
+                    return AIMessage(content=message.content or "")
+                    
+            except ImportError:
+                # Fallback: create a simple object with content attribute
+                class SimpleMessage:
+                    def __init__(self, content, tool_calls=None):
+                        self.content = content
+                        self.tool_calls = tool_calls or []
+                
+                if message.tool_calls:
+                    tool_calls = []
+                    for tc in message.tool_calls:
+                        tool_calls.append({
+                            "name": tc.function.name,
+                            "args": json.loads(tc.function.arguments),
+                            "id": tc.id
+                        })
+                    return SimpleMessage(message.content or "", tool_calls)
+                else:
+                    return SimpleMessage(message.content or "")
+            
+        except Exception as e:
+            logger.error(f"Error in chat completion with message response: {e}")
+            raise
+    
     async def _handle_tool_calls(self, assistant_message, original_messages: List[Dict[str, str]]) -> str:
         """Handle tool calls from the assistant"""
         # Add assistant message with tool calls to conversation

{isa_model-0.2.8.dist-info → isa_model-0.3.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: isa_model
-Version: 0.2.8
+Version: 0.3.0
 Summary: Unified AI model serving framework
 Author: isA_Model Contributors
 Classifier: Development Status :: 3 - Alpha

{isa_model-0.2.8.dist-info → isa_model-0.3.0.dist-info}/RECORD

@@ -40,7 +40,7 @@ isa_model/inference/services/embedding/openai_embed_service.py,sha256=47DEQpj8HB
 isa_model/inference/services/llm/__init__.py,sha256=C6t9w33j3Ap4oGcJal9-htifKe0rxwws_kC3F-_B_Ps,341
 isa_model/inference/services/llm/base_llm_service.py,sha256=hf4egO9_s3rOQYwyhDS6O_8ECIAltkj4Ir89PTosraE,8381
 isa_model/inference/services/llm/ollama_llm_service.py,sha256=EfLdoovyrChYBlGreQukpSZt5l6DkfXwjjmPPovmm70,12934
-isa_model/inference/services/llm/openai_llm_service.py,sha256=nCmWSr8dJKyiigHXRBMSqMQN9c-UYnzgNR4cWPgo2K4,14196
+isa_model/inference/services/llm/openai_llm_service.py,sha256=XarEWzPg3DnITxrhkVtdR1RC0puklFIAUALgC61P8LM,19279
 isa_model/inference/services/llm/triton_llm_service.py,sha256=ZFo7JoZ799Nvyi8Cz1jfWOa6TUn0hDRJtBrotadMAd4,17673
 isa_model/inference/services/ml/base_ml_service.py,sha256=mLBA6ENowa3KVzNqHyhWxf_Pr-cJJj84lDE4TniPzYI,2894
 isa_model/inference/services/ml/sklearn_ml_service.py,sha256=Lf9JrwvI25lca7JBbjB_e66eAUtXFbwxZ3Hs13dVGkA,5512
@@ -80,7 +80,7 @@ isa_model/training/core/config.py,sha256=oqgKpBvtzrN6jwLIQYQ2707lH6nmjrktRiSxp9i
 isa_model/training/core/dataset.py,sha256=XCFsnf0NUMU1dJpdvo_CAMyvXB-9_RCUEiy8TU50e20,7802
 isa_model/training/core/trainer.py,sha256=h5TjqjdFr0Fsv5y4-0siy1KmOlqLfliVaUXybvuoeXU,26932
 isa_model/training/core/utils.py,sha256=Nik0M2ssfNbWqP6fKO0Kfyhzr_H6Q19ioxB-qCYbn5E,8387
-isa_model-0.2.8.dist-info/METADATA,sha256=tIhw9WG4m-MOxkc20hn2wqSwuh0T8PlelyM-Q63kkVE,12226
-isa_model-0.2.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-isa_model-0.2.8.dist-info/top_level.txt,sha256=eHSy_Xb3kNkh2kK11mi1mZh0Wz91AQ5b8k2KFYO-rE8,10
-isa_model-0.2.8.dist-info/RECORD,,
+isa_model-0.3.0.dist-info/METADATA,sha256=vKAOkCdWjst6VFeisv1QxEHUEzxJpMOd5FO-RMG_C6M,12226
+isa_model-0.3.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+isa_model-0.3.0.dist-info/top_level.txt,sha256=eHSy_Xb3kNkh2kK11mi1mZh0Wz91AQ5b8k2KFYO-rE8,10
+isa_model-0.3.0.dist-info/RECORD,,