1、修改api文件位置

2、意图识别继承langfuse

.env

@@ -1,11 +1,10 @@
-OPENAI_API_KEY=sk-xxaiabmfhzwwpijuledllkmkzhzwsqeicjxmjwnvriqpwmpk
 OPENAI_API_BASE=https://api.siliconflow.cn/v1/
 MODEL_NAME=deepseek-ai/DeepSeek-V3
-MINI_MODEL_NAME=Qwen/Qwen2.5-72B-Instruct-128K
+
+XINFERENCE_URL=http://10.1.16.39:9995
 RERANKER_MODEL_NAME=bge-reranker-v2-m3
 EMBEDDING_MODEL_NAME=bge-m3
 
-XINFERENCE_URL=http://10.1.16.39:9995
 
 DIFY_BSAE_URL=http://10.1.16.39/v1
 DIFY_APP_KEY=app-CPoOMaGDsLRPAe9TW7Xjhszy
@@ -15,4 +14,10 @@ DIFY_PG_HOST = 10.1.16.39
 DIFY_PG_PORT = 5432
 DIFY_PG_USER = postgres
 DIFY_PG_PASSWORD = difyai123456
-DIFY_PG_DATABASE = dify
+DIFY_PG_DATABASE = dify
+
+
+ENABLE_LANGFUSE=true
+LANGFUSE_PUBLIC_KEY=pk-lf-4e9b7cbe-528c-4697-b73c-33257a60072c
+LANGFUSE_SECRET_KEY=sk-lf-cd8a78c5-2538-455e-a85a-87b6e1aa69d0
+LANGFUSE_HOST=http://10.1.6.34:3000

pyproject.toml

@@ -32,8 +32,6 @@ dependencies = [
     "xlsxwriter>=3.2.5",
 ]
 
-[[project.source]]
+[[tool.uv.index]]
-name = "aliyun"
+url = "https://mirrors.aliyun.com/pypi/simple"
-url = "https://mirrors.aliyun.com/pypi/simple/"
+default = true
-type = "index"
-priority = 1

rag2_0/api/DifyQueryRetrieval_api.py

@@ -44,6 +44,7 @@ class RetrieveRequest(BaseModel):
     data_set_list: str
     query_expand_dict: dict | str = Field(default="{}")
     topk: int = Field(default=4)
+    metadata_filtering_conditions : dict = Field(default={})
 
 # 创建FastAPI应用
 app = FastAPI(
@@ -103,7 +104,8 @@ async def retrieve(request: RetrieveRequest):
             query_list, 
             data_set_list,
             query_expand_dict=query_expand_dict,
-            top_k=request.topk
+            top_k=request.topk,
+            metadata_filtering_conditions=request.metadata_filtering_conditions
         )
         end_time = time.time()
         
@@ -116,11 +118,11 @@ async def retrieve(request: RetrieveRequest):
 if __name__ == "__main__":
     # 使用Uvicorn运行FastAPI应用
     import uvicorn
-    uvicorn.run("rag2_0.dify.DifyQueryRetrieval_api:app", host="0.0.0.0", port=8002, reload=False, workers=1, log_level="info") 
+    uvicorn.run("rag2_0.api.DifyQueryRetrieval_api:app", host="0.0.0.0", port=9002, reload=False, workers=1, log_level="info") 
     # # 使用uvicorn启动服务
     # import uvicorn
     # uvicorn.run(
-    #     "rag2_0.dify.intent_recognition_api:app", 
+    #     "rag2_0.api.DifyQueryRetrieval_api:app", 
     #     host="0.0.0.0", 
     #     port=8001, 
     #     reload=False,  # 开发环境启用热重载

rag2_0/dify/DifyQueryRetrieval.py

@@ -23,12 +23,13 @@ class DifyQueryRetrieval:
         datasets_json = datasets.json()
         return {dataset["name"]:dataset for dataset in datasets_json["data"]}
 
-    def retrieve_by_dataset(self, query: str, dataset_name: str) -> Dict[str, Any]:
+    def retrieve_by_dataset(self, query: str, dataset_name: str, metadata_filtering_conditions:dict = {}) -> Dict[str, Any]:
         try:
             dataset_id = self._datasets_list[dataset_name]["id"]
             retrieval_model = self._datasets_list[dataset_name]["retrieval_model_dict"]
             knowledge_base_client = KnowledgeBaseClient(api_key=self._dify_dataset_key, base_url=self._dify_base_url, dataset_id=dataset_id)
-            
+            if len(metadata_filtering_conditions) !=0:
+                retrieval_model["metadata_filtering_conditions"]=metadata_filtering_conditions
             documents = knowledge_base_client.retrieve(query, retrieval_model=retrieval_model, timeout=300)
             retrieved_documents = documents.json().get("records", [])
             
@@ -51,7 +52,7 @@ class DifyQueryRetrieval:
                 "documents": []
             }
 
-    async def retrieve_by_dataset_async(self, query: str, dataset_name: str) -> Dict[str, Any]:
+    async def retrieve_by_dataset_async(self, query: str, dataset_name: str, metadata_filtering_conditions:dict = {}) -> Dict[str, Any]:
         """
         异步版本的retrieve_by_dataset方法
         
@@ -67,7 +68,8 @@ class DifyQueryRetrieval:
             return await asyncio.to_thread(
                 self.retrieve_by_dataset,
                 query,
-                dataset_name
+                dataset_name,
+                metadata_filtering_conditions
             )
         except Exception as e:
             logging.error(f"异步检索数据集 {dataset_name} 时出错: {str(e)}", exc_info=True)
@@ -77,7 +79,13 @@ class DifyQueryRetrieval:
                 "documents": []
             }
 
-    async def retrieve_api_async(self, original_query: str, query_list: List[str], data_set_list: List[str], query_expand_dict: dict, top_k: int = 5)->Dict[str, Any]:
+    async def retrieve_api_async(self, 
+                                original_query: str, 
+                                query_list: List[str], 
+                                data_set_list: List[str], 
+                                query_expand_dict: dict, 
+                                top_k: int = 5,
+                                metadata_filtering_conditions:dict = {})->Dict[str, Any]:
         """
         异步版本的retrieve_api方法，使用asyncio代替线程池
         
@@ -105,7 +113,7 @@ class DifyQueryRetrieval:
                     continue
                 
                 # 创建异步任务
-                task = self.retrieve_by_dataset_async(query, dataset)
+                task = self.retrieve_by_dataset_async(query, dataset, metadata_filtering_conditions)
                 tasks.append(task)
         
         # 并发执行所有异步任务

rag2_0/intent_recognition/IntentRecognition.py

@@ -19,6 +19,8 @@ import jieba
 import time
 import threading
 
+from langchain_openai import ChatOpenAI
+
 from .PromptTemplates import (classification_prompt, query_rewrite_prompt_pro, 
                               extract_nouns_prompt, classification_info, 
                               slot_filling_prompt, step_back_prompt, 
@@ -32,7 +34,10 @@ from .DataModels import (
     StepBackPrompt, HypotheticalDocument
 )
 from .ProfessionalNounVector import ProfessionalNounRetriever, AsyncProfessionalNounRetriever
-from rag2_0.tool.ModelTool import XinferenceReRankerModel, OpenAiLLM
+from rag2_0.tool.APIKeyManager import APIKeyManager
+
+TEMPERATURE = 0.4
+TOP_P = 0.7
 
 class AsyncIntentRecognizer:
     SOFT_WIKI_PATH = "data/wiki_data"
@@ -59,21 +64,6 @@ class AsyncIntentRecognizer:
             model_name: 要使用的模型名称
             vector_index_dir: 向量索引目录，如果为None则使用默认目录
         """
-        api_key = os.getenv("OPENAI_API_KEY")
-        base_url = os.getenv("OPENAI_API_BASE")
-        model_name = os.getenv("MODEL_NAME", "gpt-3.5-turbo")
-        # 初始化LLM
-        llm_params = {
-            "temperature": 0.4,  # 降低随机性，使结果更确定
-            "top_p": 0.7,
-            "model": model_name,
-            "api_key": api_key,
-            "base_url": base_url
-        }
-        
-        self._llm = OpenAiLLM(**llm_params) 
-        llm_params["model"] = os.getenv("MINI_MODEL_NAME", "gpt-3.5-turbo")
-        self._llm_mini = OpenAiLLM(**llm_params) 
         
         # 加载suffix关键词
         self._suffix_keywords = self._load_suffix_keywords()
@@ -199,7 +189,15 @@ class AsyncIntentRecognizer:
         # 解析输出
         try:
             # 异步调用LLM
-            response = await self._llm.invoke_async(formatted_prompt, False)
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+            llm.with_structured_output(Classification)
+            response = await llm.ainvoke(formatted_prompt)
 
             # 尝试直接解析JSON响应
             response.content = response.content.strip()
@@ -266,8 +264,17 @@ class AsyncIntentRecognizer:
             terms_list_parser = PydanticOutputParser(pydantic_object=TermList)
             formatted_prompt = formatted_prompt.replace("{output_format}", terms_list_parser.get_format_instructions())
             
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+            llm.with_structured_output(TermList)
+
             # 异步调用LLM
-            response = await self._llm.invoke_async(formatted_prompt, False)
+            response = await llm.ainvoke(formatted_prompt)
             
             # 尝试使用Pydantic解析器解析TermList
             response.content = response.content.strip()
@@ -349,7 +356,16 @@ class AsyncIntentRecognizer:
         """
 
         try:
-            response = await self._llm.invoke_async(prompt, False, response_format={"type": "json_object"})
+
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+                
+            response = await llm.ainvoke(prompt, response_format={"type": "json_object"})
             response.content = response.content.strip()
             clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
             parsed_output = JsonOutputParser().parse(clean_output)
@@ -389,8 +405,17 @@ class AsyncIntentRecognizer:
                                                          context=context)
         # 解析输出
         try:
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+            llm.with_structured_output(QueryRewrite)
+
             # 异步调用LLM
-            response = await self._llm.invoke_async(formatted_prompt, False)
+            response = await llm.ainvoke(formatted_prompt)
             response.content = response.content.strip()
             clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
             parsed_output = query_rewrite_parser.parse(clean_output)
@@ -634,8 +659,18 @@ class AsyncIntentRecognizer:
             previous_slots=json.dumps(previous_slots,ensure_ascii=False),
         )
         try:
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+            llm.with_structured_output(slot_model_class)
+
+
             # 异步调用LLM
-            response = await self._llm.invoke_async(formatted_prompt, False)
+            response = await llm.ainvoke(formatted_prompt)
             response.content = response.content.strip()
             clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
             # 尝试解析LLM响应
@@ -669,10 +704,17 @@ class AsyncIntentRecognizer:
         )
         
         try:
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P
+            )
+            llm.with_structured_output(StepBackPrompt)
+
             # 异步调用LLM
-            response = await self._llm.invoke_async(formatted_prompt, False)
+            response = await llm.ainvoke(formatted_prompt)
-            
-            # 解析输出
             response.content = response.content.strip()
             clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
             parsed_output = step_back_parser.parse(clean_output)
@@ -728,9 +770,18 @@ class AsyncIntentRecognizer:
         """
         
         try:
+
+            llm = ChatOpenAI(
+                api_key=APIKeyManager.get_api_key(),
+                openai_api_base=os.getenv("OPENAI_API_BASE"),
+                model_name=os.getenv("MODEL_NAME"),
+                temperature=TEMPERATURE,
+                top_p=TOP_P,
+            )
+
             # 异步调用LLM
             start_time = time.time()
-            response = await self._llm.invoke_async(prompt, False, response_format={"type": "json_object"})
+            response = await llm.ainvoke(prompt, response_format={"type": "json_object"})
             end_time = time.time()
             
             # 解析JSON响应

start_AnswerType.sh

@@ -10,7 +10,7 @@ else
     # 启动一个名为AnswerType的screen会话，并在其中执行后续命令
     screen -dmS AnswerType bash -c "
         cd \"$SCRIPT_DIR\"
-        uv run uvicorn rag2_0.dify.AnswerType:app --host 0.0.0.0 --port 8003 --workers 1 
+        uv run uvicorn rag2_0.api.AnswerType_api:app --host 0.0.0.0 --port 8003 --workers 1 
     "
 
     # 输出提示信息

start_DifyQueryRetrieval_api.sh

@@ -10,7 +10,7 @@ else
     # 启动一个名为DifyQueryRetrieval_api的screen会话，并在其中执行后续命令
     screen -dmS DifyQueryRetrieval_api bash -c "
         cd \"$SCRIPT_DIR\"
-        uv run uvicorn rag2_0.dify.DifyQueryRetrieval_api:app --host 0.0.0.0 --port 8002 --workers 25
+        uv run uvicorn rag2_0.api.DifyQueryRetrieval_api:app --host 0.0.0.0 --port 8002 --workers 25
     "
 
     # 输出提示信息

start_intent_recognition_api.sh

@@ -10,7 +10,7 @@ else
     # 启动一个名为intent_recognition_api的screen会话，并在其中执行后续命令
     screen -dmS intent_recognition_api bash -c "
         cd \"$SCRIPT_DIR\"
-        uv run uvicorn rag2_0.dify.intent_recognition_api:app --host 0.0.0.0 --port 8001 --workers 25
+        uv run uvicorn rag2_0.api.intent_recognition_api:app --host 0.0.0.0 --port 8001 --workers 25
     "
 
     # 输出提示信息