dsProject/dsRag/Start.py

import asyncio
from contextlib import asynccontextmanager
from logging.handlers import RotatingFileHandler

import jieba  # 导入 jieba 分词库
import uvicorn
from fastapi import FastAPI, Request, HTTPException
from pydantic import BaseModel, Field, ValidationError
from fastapi.staticfiles import StaticFiles
from openai import OpenAI
from sse_starlette.sse import EventSourceResponse
from gensim.models import KeyedVectors
from Config import Config
from Config.Config import MS_MODEL_PATH, MS_MODEL_LIMIT, MS_HOST, MS_PORT, MS_MAX_CONNECTIONS, MS_NPROBE, DEEPSEEK_API_KEY, DEEPSEEK_URL, MS_COLLECTION_NAME
from Milvus.Utils.MilvusCollectionManager import MilvusCollectionManager
from Milvus.Utils.MilvusConnectionPool import *
from Milvus.Utils.MilvusConnectionPool import MilvusConnectionPool

# 初始化日志
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
handler = RotatingFileHandler('Logs/start.log', maxBytes=1024 * 1024, backupCount=5)
handler.setFormatter(logging.Formatter('%(asctime)s - %(levelname)s - %(message)s'))
logger.addHandler(handler)

# 1. 加载预训练的 Word2Vec 模型
model_path = MS_MODEL_PATH  # 替换为你的 Word2Vec 模型路径
model = KeyedVectors.load_word2vec_format(model_path, binary=False, limit=MS_MODEL_LIMIT)
logger.info(f"模型加载成功，词向量维度: {model.vector_size}")


@asynccontextmanager
async def lifespan(app: FastAPI):
    # 初始化Milvus连接池
    app.state.milvus_pool = MilvusConnectionPool(host=MS_HOST, port=MS_PORT, max_connections=MS_MAX_CONNECTIONS)

    # 初始化集合管理器
    app.state.collection_manager = MilvusCollectionManager(MS_COLLECTION_NAME)
    app.state.collection_manager.load_collection()

    # 初始化DeepSeek客户端
    app.state.deepseek_client = OpenAI(
        api_key=Config.DEEPSEEK_API_KEY,
        base_url=Config.DEEPSEEK_URL
    )
    yield

    # 关闭Milvus连接池
    app.state.milvus_pool.close()


app = FastAPI(lifespan=lifespan)

# 挂载静态文件目录
app.mount("/static", StaticFiles(directory="Static"), name="static")


# 将文本转换为嵌入向量
def text_to_embedding(text):
    words = jieba.lcut(text)  # 使用 jieba 分词
    print(f"文本: {text}, 分词结果: {words}")
    embeddings = [model[word] for word in words if word in model]
    logger.info(f"有效词向量数量: {len(embeddings)}")
    if embeddings:
        avg_embedding = sum(embeddings) / len(embeddings)
        logger.info(f"生成的平均向量: {avg_embedding[:5]}...")  # 打印前 5 维
        return avg_embedding
    else:
        logger.warning("未找到有效词，返回零向量")
        return [0.0] * model.vector_size


async def generate_stream(client, milvus_pool, collection_manager, query):
    """生成SSE流"""
    # 从连接池获取连接
    connection = milvus_pool.get_connection()

    try:
        # 1. 将查询文本转换为向量
        current_embedding = text_to_embedding(query)

        # 2. 搜索相关数据
        search_params = {
            "metric_type": "L2",  # 使用 L2 距离度量方式
            "params": {"nprobe": MS_NPROBE}  # 设置 IVF_FLAT 的 nprobe 参数
        }
        # 7. 将文本转换为嵌入向量
        results = collection_manager.search(current_embedding, search_params, limit=5)  # 返回 2 条结果

        # 3. 处理搜索结果
        logger.info("最相关的历史对话:")
        context = ""
        if results:
            for hits in results:
                for hit in hits:
                    try:
                        # 查询非向量字段
                        record = collection_manager.query_by_id(hit.id)
                        logger.info(f"ID: {hit.id}")
                        logger.info(f"标签: {record['tags']}")
                        logger.info(f"用户问题: {record['user_input']}")
                        
                        # 获取完整内容
                        full_content = record['tags'].get('full_content', record['user_input'])
                        context = context + full_content
                        
                        logger.info(f"时间: {record['timestamp']}")
                        logger.info(f"距离: {hit.distance}")
                        logger.info("-" * 40)  # 分隔线
                    except Exception as e:
                        logger.error(f"查询失败: {e}")
        else:
            logger.warning("未找到相关历史对话，请检查查询参数或数据。")

        prompt = f"""
信息检索与回答助手
根据以下关于'{query}'的相关信息：

基本信息
- 语言: 中文
- 描述: 根据提供的材料检索信息并回答问题
- 特点: 快速准确提取关键信息，清晰简洁地回答

相关信息
{context}

回答要求
1. 准确无误，来源可靠
2. 客观公正，避免主观判断
3. 使用HTML格式返回，包含适当的段落、列表和标题标签
4. 确保内容结构清晰，便于前端展示
"""

        response = client.chat.completions.create(
            model="deepseek-chat",
            messages=[
                {"role": "system", "content": "你是一个专业的文档整理助手"},
                {"role": "user", "content": prompt}
            ],
            temperature=0.3,
            stream=False
        )

        yield {"data": response.choices[0].message.content}
    except Exception as e:
        yield {"data": f"生成报告时出错: {str(e)}"}
    finally:
        # 释放连接
        milvus_pool.release_connection(connection)


"""
http://10.10.21.22:8000/static/ai.html
小学数学中有哪些模型？
"""


class QueryRequest(BaseModel):
    query: str = Field(..., description="用户查询的问题")

@app.post("/api/rag")
async def rag_stream(request: Request):
    try:
        data = await request.json()
        query_request = QueryRequest(**data)
    except ValidationError as e:
        logger.error(f"请求体验证失败: {e.errors()}")
        raise HTTPException(status_code=422, detail=e.errors())
    except Exception as e:
        logger.error(f"请求解析失败: {str(e)}")
        raise HTTPException(status_code=400, detail="无效的请求格式")
    """RAG+DeepSeek接口"""
    async for chunk in generate_stream(
        request.app.state.deepseek_client,
        request.app.state.milvus_pool,
        request.app.state.collection_manager,
        query_request.query
    ):
        return chunk


if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)