'commit'

dsRagAnything/T3_Start.py

@@ -38,8 +38,25 @@ logger.addHandler(console_handler)
 
 
 async def lifespan(app: FastAPI):
-    # 初始化RAG
-    app.state.rag = await initialize_rag(working_dir="./Topic/Math")
+    # 初始化RAG组件
+    llm_model_func = create_llm_model_func()
+    vision_model_func = create_vision_model_func(llm_model_func)
+    embedding_func = create_embedding_func()
+
+    lightrag_instance = LightRAG(
+        working_dir=WORKING_PATH,
+        llm_model_func=llm_model_func,
+        embedding_func=embedding_func
+    )
+
+    await lightrag_instance.initialize_storages()
+    await initialize_pipeline_status()
+
+    # 创建RAG实例并保存到app.state
+    app.state.rag = RAGAnything(
+        lightrag=lightrag_instance,
+        vision_model_func=vision_model_func,
+    )
     yield
 
     # 清理资源
@@ -58,7 +75,7 @@ app = FastAPI(lifespan=lifespan)
 app.mount("/static", StaticFiles(directory="Static"), name="static")
 
 # 数学索引文件目录
-#WORKING_PATH = "./Topic/Math"
+# WORKING_PATH = "./Topic/Math"
 
 # 苏轼索引文件目录
 WORKING_PATH = "./Topic/Chinese"
@@ -68,35 +85,18 @@ WORKING_PATH = "./Topic/Chinese"
 async def rag(request: fastapi.Request):
     data = await request.json()
     query = data.get("query")
+    # 关闭参考资料
+    query = query + "\n 不要输出参考资料！"
 
     async def generate_response_stream(query: str):
         try:
-            llm_model_func = create_llm_model_func()
-            vision_model_func = create_vision_model_func(llm_model_func)
-            embedding_func = create_embedding_func()
-
-            lightrag_instance = LightRAG(
-                working_dir=WORKING_PATH,
-                llm_model_func=llm_model_func,
-                embedding_func=embedding_func
-            )
-
-            await lightrag_instance.initialize_storages()
-            await initialize_pipeline_status()
-
-            rag = RAGAnything(
-                lightrag=lightrag_instance,
-                vision_model_func=vision_model_func,
-            )
-
-            # 使用stream=True参数确保流式输出
-            resp = await rag.aquery(
+            # 直接使用app.state中已初始化的rag实例
+            resp = await app.state.rag.aquery(
                 query=query,
-                mode="hybrid",  # 直接传入mode参数
-                stream=True  # 直接传入stream参数
+                mode="hybrid",
+                stream=True
             )
 
-            # 直接处理流式响应
             async for chunk in resp:
                 if not chunk:
                     continue

dsRagAnything/Topic/Chinese/kv_store_llm_response_cache.json

@@ -2201,6 +2201,16 @@
       "embedding_min": null,
       "embedding_max": null,
       "original_prompt": "苏轼的家人都有谁？"
+    },
+    "585eef85444b748e7c7fce9281f7152b": {
+      "return": "{\"high_level_keywords\": [\"\\u82cf\\u8f7c\", \"\\u5bb6\\u5ead\\u6210\\u5458\", \"\\u5bb6\\u65cf\\u5173\\u7cfb\"], \"low_level_keywords\": [\"\\u7a0b\\u592b\\u4eba\", \"\\u738b\\u5f17\", \"\\u82cf\\u8f99\", \"\\u82cf\\u8fc8\", \"\\u7709\\u5c71\"]}",
+      "cache_type": "keywords",
+      "chunk_id": null,
+      "embedding": null,
+      "embedding_shape": null,
+      "embedding_min": null,
+      "embedding_max": null,
+      "original_prompt": "苏轼的家人都有谁？\n 不要输出参考资料！"
     }
   }
 }