ДОбился запуска

.gitignore

@@ -0,0 +1,6 @@
+__pycache__
+agent_env
+.lock
+.sqlite
+qdrant_data
+qdrant_storage

agent_core.py

@@ -1,15 +1,13 @@
 from typing import TypedDict, List, Optional, Annotated
 import operator
 from langchain_community.llms import Ollama
-from langchain_community.embeddings import OllamaEmbeddings
 from langchain_community.vectorstores import Qdrant
-from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_community.document_loaders import DirectoryLoader, PyPDFLoader
 from langgraph.graph import StateGraph, END
 from langgraph.graph.message import add_messages
-from langchain_core.messages import HumanMessage, AIMessage, SystemMessage
+from langchain_core.messages import AIMessage, SystemMessage
 from langchain.tools import tool
-from langchain.agents import AgentExecutor, create_react_agent
+# from langchain.agents import AgentExecutor, create_react_agent
 import numpy as np
 
 # Определяем состояние агента
@@ -22,16 +20,11 @@ class AgentState(TypedDict):
 
 # Инициализация LLM через Ollama
 llm = Ollama(
-    model="llama3.2:3b",
+    model="qwen2.5-coder-16k:14b",
     temperature=0.1,  # Низкая температура для консистентности
     num_predict=1024,  # Максимальная длина ответа
 )
 
-# Инициализация эмбеддингов
-embeddings = OllamaEmbeddings(
-    model="nomic-embed-text",  # Хорошие локальные эмбеддинги
-)
-
 # Создаем инструменты для агента
 @tool
 def search_knowledge_base(query: str) -> str:
@@ -127,6 +120,3 @@ def create_agent_graph():
     workflow.add_edge("generate", END)
     
     return workflow.compile()
-
-# Создаем и компилируем граф
-agent_graph = create_agent_graph()

agentic_rag_interface.py

@@ -1,6 +1,10 @@
 import gradio as gr
 import asyncio
+import os
 from typing import List
+from agent_core import create_agent_graph
+from local_knowledge_base import LocalKnowledgeBase
+from langchain_core.messages import HumanMessage
 
 class AgenticRAGInterface:
     def __init__(self, agent_graph, knowledge_base):
@@ -61,8 +65,15 @@ class AgenticRAGInterface:
         self.conversation_history = []
         return "История очищена", ""
 
+# Инициализируем базу
+kb = LocalKnowledgeBase()
+# Загружаем документы (если есть)
+if os.path.exists("./documents"):
+    kb.load_documents("./documents")
+
+
 # Создаем интерфейс
-interface = AgenticRAGInterface(agent_graph, kb)
+interface = AgenticRAGInterface(create_agent_graph(), kb)
 
 # Создаем Gradio интерфейс
 def create_gradio_interface():

local_knowledge_base.py

@@ -1,19 +1,28 @@
 import os
 from qdrant_client import QdrantClient
 from qdrant_client.models import Distance, VectorParams, PointStruct
-from langchain_community.document_loaders import TextLoader, PyPDFLoader, DirectoryLoader
-from langchain.text_splitter import Language
+from langchain_community.document_loaders import (
+    TextLoader,
+    PyPDFLoader,
+    DirectoryLoader,
+)
+from langchain_text_splitters import Language, RecursiveCharacterTextSplitter
+from langchain_community.embeddings import OllamaEmbeddings
+
 
 class LocalKnowledgeBase:
     def __init__(self, collection_name="documents", persist_dir="./qdrant_data"):
         # Инициализируем клиент Qdrant
         self.client = QdrantClient(
-            path=persist_dir,  # Локальное хранение
-            prefer_grpc=True
+            path=persist_dir, prefer_grpc=True  # Локальное хранение
         )
 
         self.collection_name = collection_name
-        self.embeddings = embeddings
+
+        # Инициализация эмбеддингов
+        self.embeddings = OllamaEmbeddings(
+            model="nomic-embed-text",  # Хорошие локальные эмбеддинги
+        )
 
         # Создаем коллекцию если её нет
         self._create_collection()
@@ -28,22 +37,21 @@ class LocalKnowledgeBase:
                 collection_name=self.collection_name,
                 vectors_config=VectorParams(
                     size=768,  # Размерность эмбеддингов nomic-embed-text
-                    distance=Distance.COSINE
-                )
+                    distance=Distance.COSINE,
+                ),
             )
             print(f"Создана коллекция {self.collection_name}")
 
     def load_documents(self, directory_path: str):
         """Загрузка документов из директории"""
         loaders = {
-            '.pdf': PyPDFLoader,
-            '.txt': lambda path: DirectoryLoader(path, glob="**/*.txt"),
-            '.docx': lambda path: DirectoryLoader(path, glob="**/*.docx"),
-            '.hpp': lambda path: TextLoader(path, encoding='utf-8'),
-            '.cpp': lambda path: TextLoader(path, encoding='utf-8'),   # добавить
-            '.h': lambda path: TextLoader(path, encoding='utf-8'),     # добавить
-            '.cc': lambda path: TextLoader(path, encoding='utf-8'),    # добавить
-            
+            ".pdf": PyPDFLoader,
+            ".txt": lambda path: DirectoryLoader(path, glob="**/*.txt"),
+            ".docx": lambda path: DirectoryLoader(path, glob="**/*.docx"),
+            ".hpp": lambda path: TextLoader(path, encoding="utf-8"),
+            ".cpp": lambda path: TextLoader(path, encoding="utf-8"),  # добавить
+            ".h": lambda path: TextLoader(path, encoding="utf-8"),  # добавить
+            ".cc": lambda path: TextLoader(path, encoding="utf-8"),  # добавить
         }
 
         all_documents = []
@@ -65,7 +73,7 @@ class LocalKnowledgeBase:
             language=Language.CPP,
             chunk_size=1000,
             chunk_overlap=200,
-            length_function=len
+            length_function=len,
         )
 
         chunks = text_splitter.split_documents(all_documents)
@@ -91,26 +99,20 @@ class LocalKnowledgeBase:
                     "text": doc.page_content,
                     "source": doc.metadata.get("source", "unknown"),
                     "page": doc.metadata.get("page", 0),
-                    "file_type": "cpp"
-                }
+                    "file_type": "cpp",
+                },
             )
             points.append(point)
 
             # Пакетная загрузка каждые 100 точек
             if len(points) >= 100:
-                self.client.upsert(
-                    collection_name=self.collection_name,
-                    points=points
-                )
+                self.client.upsert(collection_name=self.collection_name, points=points)
                 points = []
                 print(f"Индексировано {i+1} документов")
 
         # Загружаем оставшиеся
         if points:
-            self.client.upsert(
-                collection_name=self.collection_name,
-                points=points
-            )
+            self.client.upsert(collection_name=self.collection_name, points=points)
 
         print(f"Индексация завершена. Всего документов: {len(documents)}")
 
@@ -120,26 +122,22 @@ class LocalKnowledgeBase:
         query_embedding = self.embeddings.embed_query(query)
 
         # Ищем в Qdrant
-        search_result = self.client.search(
+        search_result = self.client.query_points(
             collection_name=self.collection_name,
-            query_vector=query_embedding,
-            limit=top_k
+            query=query_embedding,
+            limit=top_k,
         )
 
         # Форматируем результаты
         results = []
-        for hit in search_result:
-            results.append({
-                "text": hit.payload["text"],
-                "score": hit.score,
-                "source": hit.payload.get("source", "unknown")
-            })
+        for hit in search_result.points:
+            results.append(
+                {
+                    "text": hit.payload["text"],
+                    "score": hit.score,
+                    "source": hit.payload.get("source", "unknown"),
+                }
+            )
 
         return results
 
-# Инициализируем базу знаний
-kb = LocalKnowledgeBase()
-
-# Загружаем документы (если есть)
-if os.path.exists("./documents"):
-    kb.load_documents("./documents")