以下是一个完整的、可运行的 Python 实现的 RAG（Retrieval-Augmented Generation）程序，具备以下功能： --- ### ✅ 功能说明： - 从本地 Ollama 获取模型列表 - 使用本地 SentenceTransformer 模型进行文档嵌入 - 使用 FAISS 构建向量数据库进行语义检索 - 支持参数控制（Top-K、Temperature、Max Tokens） - 使用 PyQt5 构建图形界面 - 支持文档加载与处理（使用 `langchain` 和 `unstructured`） --- ### 📁 项目结构（简化） ``` rag_app/ ├── main.py ├── documents/ │ └── sample.txt └── requirements.txt ``` --- ### 📦 安装依赖（`requirements.txt`） ```txt requests sentence-transformers faiss-cpu langchain unstructured PyQt5 ``` --- ### 🧠 `main.py`（完整代码） ```python import sys import os import requests import numpy as np from PyQt5.QtWidgets import QApplication, QWidget, QVBoxLayout, QLabel, QLineEdit, QPushButton, QTextEdit, QComboBox from sentence_transformers import SentenceTransformer from langchain.document_loaders import TextLoader from langchain.text_splitter import RecursiveCharacterTextSplitter import faiss # ========== Ollama 模型获取 ========== def get_ollama_models(): try: response = requests.get("http://localhost:11434/api/tags") if response.status_code == 200: return [model["name"] for model in response.json()["models"]] else: return ["连接失败"] except Exception as e: print("Ollama连接失败:", e) return ["连接失败"] # ========== 文档加载与处理 ========== def load_documents(folder_path="documents"): documents = [] for filename in os.listdir(folder_path): if filename.endswith(".txt"): loader = TextLoader(os.path.join(folder_path, filename)) documents.extend(loader.load()) text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100) return text_splitter.split_documents(documents) # ========== 嵌入模型加载 ========== embedding_model = SentenceTransformer("all-MiniLM-L6-v2") # ========== 构建 FAISS 索引 ========== texts = [doc.page_content for doc in load_documents()] embeddings = embedding_model.encode(texts) dimension = embeddings.shape[1] index = faiss.IndexFlatL2(dimension) index.add(np.array(embeddings)) # ========== 检索函数 ========== def retrieve_documents(query, top_k=5): query_embedding = embedding_model.encode([query]) distances, indices = index.search(np.array(query_embedding), top_k) return [texts[i] for i in indices[0]] # ========== 生成回答函数 ========== def generate_answer(prompt, model_name, temperature=0.7, max_tokens=200): url = "http://localhost:11434/api/generate" payload = { "model": model_name, "prompt": prompt, "stream": False, "temperature": temperature, "max_tokens": max_tokens, } response = requests.post(url, json=payload) if response.status_code == 200: return response.json()["response"] else: return "生成失败：" + response.text # ========== RAG 主流程 ========== def rag_query(query, model_name, top_k=5, temperature=0.7, max_tokens=200): retrieved_docs = retrieve_documents(query, top_k) context = "\n".join(retrieved_docs) full_prompt = f"Context: {context}\n\nQuestion: {query}\n\nAnswer:" answer = generate_answer(full_prompt, model_name, temperature, max_tokens) return answer # ========== PyQt5 GUI ========== class RAGApp(QWidget): def __init__(self): super().__init__() self.setWindowTitle("RAG 程序 - 本地部署") self.layout = QVBoxLayout() self.model_label = QLabel("选择模型:") self.layout.addWidget(self.model_label) self.model_combo = QComboBox() self.model_combo.addItems(get_ollama_models()) self.layout.addWidget(self.model_combo) self.query_input = QLineEdit() self.query_input.setPlaceholderText("输入你的问题...") self.layout.addWidget(self.query_input) self.top_k_input = QLineEdit() self.top_k_input.setPlaceholderText("Top-K 检索数量 (默认5)") self.layout.addWidget(self.top_k_input) self.temperature_input = QLineEdit() self.temperature_input.setPlaceholderText("Temperature (默认0.7)") self.layout.addWidget(self.temperature_input) self.max_tokens_input = QLineEdit() self.max_tokens_input.setPlaceholderText("Max Tokens (默认200)") self.layout.addWidget(self.max_tokens_input) self.submit_button = QPushButton("提交") self.submit_button.clicked.connect(self.on_submit) self.layout.addWidget(self.submit_button) self.result_output = QTextEdit() self.result_output.setReadOnly(True) self.layout.addWidget(self.result_output) self.setLayout(self.layout) def on_submit(self): query = self.query_input.text() model_name = self.model_combo.currentText() top_k = int(self.top_k_input.text()) if self.top_k_input.text() else 5 temperature = float(self.temperature_input.text()) if self.temperature_input.text() else 0.7 max_tokens = int(self.max_tokens_input.text()) if self.max_tokens_input.text() else 200 answer = rag_query(query, model_name, top_k=top_k, temperature=temperature, max_tokens=max_tokens) self.result_output.setText(answer) # ========== 启动应用 ========== if __name__ == "__main__": app = QApplication(sys.argv) window = RAGApp() window.resize(600, 500) window.show() sys.exit(app.exec_()) ``` --- ### 📌 注意事项： 1. **Ollama 必须运行在本地**，启动命令：`ollama serve` 2. 需要提前下载模型，如：`ollama pull llama3` 或 `ollama pull qwen` 3. 放入 `documents/` 文件夹中的 `.txt` 文件将被自动加载 4. 如果使用其他格式（如 PDF），可扩展 `langchain` 支持的加载器 --- ### ✅ 示例运行流程： 1. 安装依赖：`pip install -r requirements.txt` 2. 启动 Ollama：`ollama serve` 3. 下载模型：`ollama pull llama3` 4. 放入文档到 `documents/` 文件夹 5. 运行程序：`python main.py` 6. 在界面中选择模型、输入问题、调节参数、查看结果 --- ###