大模型系列-fastgpt,ollama搭建本地知识库

1. 安装fastgpt,oneapi

• docker-compose安装

1. 下载docker-compose.yml,config.json

mkdir fastgpt
cd fastgpt
curl -O https://raw.githubusercontent.com/labring/FastGPT/main/projects/app/data/config.json

# pgvector 版本(测试推荐，简单快捷)
curl -o docker-compose.yml https://raw.githubusercontent.com/labring/FastGPT/main/files/docker/docker-compose-pgvector.yml
# milvus 版本
# curl -o docker-compose.yml https://raw.githubusercontent.com/labring/FastGPT/main/files/docker/docker-compose-milvus.yml
# zilliz 版本
# curl -o docker-compose.yml https://raw.githubusercontent.com/labring/FastGPT/main/files/docker/docker-compose-zilliz.yml

2. 修改docker-compose.yml里面的内容

# 数据库的默认账号和密码仅首次运行时设置有效
# 如果修改了账号密码，记得改数据库和项目连接参数，别只改一处~
# 该配置文件只是给快速启动，测试使用。正式使用，记得务必修改账号密码，以及调整合适的知识库参数，共享内存等。
# 如何无法访问 dockerhub 和 git，可以用阿里云（阿里云没有arm包）

version: '3.3'
services:
  minio:
    container_name: minio
    image: minio/minio:RELEASE.2023-03-20T20-16-18Z
    environment:
      MINIO_ACCESS_KEY: minioadmin
      MINIO_SECRET_KEY: minioadmin
    ports:
      - '9001:9001'
      - '9000:9000'
    networks:
      - fastgpt
    volumes:
      - ./minio:/minio_data
    command: minio server /minio_data --console-address ":9001"
    healthcheck:
      test: ['CMD', 'curl', '-f', 'http://localhost:9000/minio/health/live']
      interval: 30s
      timeout: 20s
      retries: 3
  # milvus
  milvusEtcd:
    container_name: milvusEtcd
    image: quay.io/coreos/etcd:v3.5.5
    environment:
      - ETCD_AUTO_COMPACTION_MODE=revision
      - ETCD_AUTO_COMPACTION_RETENTION=1000
      - ETCD_QUOTA_BACKEND_BYTES=4294967296
      - ETCD_SNAPSHOT_COUNT=50000
    networks:
      - fastgpt
    volumes:
      - ./milvus/etcd:/etcd
    command: etcd -advertise-client-urls=http://127.0.0.1:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd
    healthcheck:
      test: ['CMD', 'etcdctl', 'endpoint', 'health']
      interval: 30s
      timeout: 20s
      retries: 3
  milvusStandalone:
    container_name: milvusStandalone
    image: milvusdb/milvus:v2.4.3
    command: ['milvus', 'run', 'standalone']
    security_opt:
      - seccomp:unconfined
    environment:
      ETCD_ENDPOINTS: milvusEtcd:2379
      MINIO_ADDRESS: minio:9000
    networks:
      - fastgpt
    volumes:
      - ./milvus/data:/var/lib/milvus
    healthcheck:
      test: ['CMD', 'curl', '-f', 'http://localhost:9091/healthz']
      interval: 30s
      start_period: 90s
      timeout: 20s
      retries: 3
    depends_on:
      - 'milvusEtcd'
      - 'minio'

  mongo:
    image: registry.cn-hangzhou.aliyuncs.com/fastgpt/mongo:5.0.18 # 阿里云
    container_name: mongo
    #restart: always
    ports:
      - 27017:27017
    networks:
      - fastgpt
    command: mongod --keyFile /data/mongodb.key --replSet rs0
    environment:
      - MONGO_INITDB_ROOT_USERNAME=myusername
      - MONGO_INITDB_ROOT_PASSWORD=mypassword
    volumes:
      - ./mongo/data:/data/db
    entrypoint:
      - bash
      - -c
      - |
        openssl rand -base64 128 > /data/mongodb.key
        chmod 400 /data/mongodb.key
        chown 999:999 /data/mongodb.key
        echo 'const isInited = rs.status().ok === 1
        if(!isInited){
          rs.initiate({
              _id: "rs0",
              members: [
                  { _id: 0, host: "mongo:27017" }
              ]
          })
        }' > /data/initReplicaSet.js
        # 启动MongoDB服务
        exec docker-entrypoint.sh "$$@" &

        # 等待MongoDB服务启动
        until mongo -u myusername -p mypassword --authenticationDatabase admin --eval "print('waited for connection')" > /dev/null 2>&1; do
          echo "Waiting for MongoDB to start..."
          sleep 2
        done

        # 执行初始化副本集的脚本
        mongo -u myusername -p mypassword --authenticationDatabase admin /data/initReplicaSet.js

        # 等待docker-entrypoint.sh脚本执行的MongoDB服务进程
        wait $$!

  # fastgpt
  sandbox:
    container_name: sandbox
    image: ghcr.io/labring/fastgpt-sandbox:v4.8.15 # git
    # image: registry.cn-hangzhou.aliyuncs.com/fastgpt/fastgpt-sandbox:v4.8.15 # 阿里云
    networks:
      - fastgpt
    #restart: always
  fastgpt:
    container_name: fastgpt
    image: ghcr.io/labring/fastgpt:v4.8.15-fix2 # git
    # image: registry.cn-hangzhou.aliyuncs.com/fastgpt/fastgpt:v4.8.15-fix2 # 阿里云
    ports:
      - 3000:3000
    networks:
      - fastgpt
    depends_on:
      - mongo
      - milvusStandalone
      - sandbox
    #restart: always
    environment:
      # 前端访问地址: http://localhost:3000
      - FE_DOMAIN=http://localhost:3000
      # root 密码，用户名为: root。如果需要修改 root 密码，直接修改这个环境变量，并重启即可。
      - DEFAULT_ROOT_PSW=1234
      # AI模型的API地址哦。务必加 /v1。这里默认填写了OneApi的访问地址。
      - OPENAI_BASE_URL=http://oneapi:3000/v1
      # AI模型的API Key。（这里默认填写了OneAPI的快速默认key，测试通后，务必及时修改）
      - CHAT_API_KEY=sk-fastgpt
      # 数据库最大连接数
      - DB_MAX_LINK=30
      # 登录凭证密钥
      - TOKEN_KEY=any
      # root的密钥，常用于升级时候的初始化请求
      - ROOT_KEY=root_key
      # 文件阅读加密
      - FILE_TOKEN_KEY=filetoken
      # MongoDB 连接参数. 用户名myusername,密码mypassword。
      - MONGODB_URI=mongodb://myusername:mypassword@mongo:27017/fastgpt?authSource=admin
      # zilliz 连接参数
      - MILVUS_ADDRESS=http://milvusStandalone:19530
      - MILVUS_TOKEN=none
      # sandbox 地址
      - SANDBOX_URL=http://sandbox:3000
      # 日志等级: debug, info, warn, error
      - LOG_LEVEL=info
      - STORE_LOG_LEVEL=warn
    volumes:
      - ./config.json:/app/data/config.json

  # oneapi
  oneapi:
    container_name: oneapi
    image: ghcr.io/songquanpeng/one-api:v0.6.7
    #image: justsong/one-api:latest
    # image: registry.cn-hangzhou.aliyuncs.com/fastgpt/one-api:v0.6.6 # 阿里云
    ports:
      - 3001:3000
    #depends_on:
    #  - mysql_my
    networks:
      - fastgpt
    #restart: always
    environment:
      # mysql 连接参数
      #- SQL_DSN=root:oneapimmysql@tcp(mysql:3306)/oneapi1
      - SQL_DSN=root:root@tcp(10.0.2.218:3306)/oneapi
      #- SQL_DSN=root:root@tcp(host.docker.internal:3306)/oneapi
      # 登录凭证加密密钥
      - SESSION_SECRET=oneapikey
      # 内存缓存
      - MEMORY_CACHE_ENABLED=true
      # 启动聚合更新，减少数据交互频率
      - BATCH_UPDATE_ENABLED=true
      # 聚合更新时长
      - BATCH_UPDATE_INTERVAL=10
      # 初始化的 root 密钥（建议部署完后更改，否则容易泄露）
      - INITIAL_ROOT_TOKEN=fastgpt
    volumes:
      - ./oneapi:/data
networks:
  fastgpt:

3. 启动容器

docker-compose up -d

4. 登录oneapi
   可以通过ip:3001访问OneAPI，默认账号为root密码为123456。
5. 登录fastgpt
   可以通过ip:3000访问FastGPT，默认账号为root密码为docker-compose.yml环境变量里设置的 DEFAULT_ROOT_PSW。

2. 安装ollama运行大模型

参考:Ollama 中文文档

2.1. 安装ollama

docker pull ollama/ollama:latest
docker run -d --gpus=all -v /media/geekplusa/GeekPlusA1/ai/models/llm/models/ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
# 使ollama保持模型加载在内存（显存）中
docker run -d --gpus=all -e OLLAMA_KEEP_ALIVE=-1  -v /media/geekplusa/GeekPlusA1/ai/models/llm/models/ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama

2.2. ollama下载模型

• 运行qwen:7b模型

docker exec -it ollama ollama run qwen:7b

• 测试

curl http://localhost:11434/api/chat -d '{
  "model": "qwen:7b",
  "messages": [
    {
      "role": "user",
      "content": "你是谁?"
    }
  ]
}'

3. 安装开源的文本向量模型

由于M3E模型不可商用,由可商用模型bge-m3替换,安装

• 下载bge-small-zh-v1.5模型

git clone https://www.modelscope.cn/Xorbits/bge-small-zh-v1.5.git
cd bge-small-zh-v1.5
wget https://www.modelscope.cn/models/Xorbits/bge-small-zh-v1.5/resolve/master/pytorch_model.bin

• 启动模型

from sentence_transformers import SentenceTransformer
import torch
from flask import Flask, request, jsonify
import os
import threading



m3e = SentenceTransformer("bge-m3")

if torch.cuda.is_available():
    m3e=m3e.to("cuda")
    print("现在使用GPU模式运行M3模型")
else:
    print("现在使用CPU模式运行M3模型")

# print("Enter your authtoken, which can be copied from https://dashboard.ngrok.com/auth")
# conf.get_default().auth_token = getpass.getpass()

os.environ["FLASK_ENV"] = "development"

app = Flask(__name__)



# Define Flask routes
@app.route("/")
def index():
    return "这是BGE-M3的API接口，请访问/v1/embeddings 使用POST请求"

@app.route('/v1/embeddings', methods=['POST'])
def embeddings():
    data = request.json
    input_text = data.get('input')
    print(f"/v1/embeddings 收到请求，输入文本为：{input_text}")
    model = data.get('model')
    if model is None:
        model = "BAAI/bge-m3"

    if input_text is None:
        return jsonify(error="No input text provided"), 400
    # input can be string or array of strings
    # if isinstance(input_text, str):
    #     sentences = [input_text]
    # elif isinstance(input_text, list):
    #     sentences = input_text
    sentences = input_text if isinstance(input_text, list) else [input_text]
    embeddings = m3e.encode(sentences,convert_to_tensor=True,show_progress_bar=True)
    embeddings = embeddings.tolist()
    data = [{"object": "embedding", "embedding": x, "index": i} for i, x in enumerate(embeddings)]

    # Mimic the response structure of the OpenAI API
    response = {
        "object": "list",
        "data": data,
        "model": model,
        "usage": {
            "prompt_tokens": 0,
            "total_tokens": 0
        }
    }

    return jsonify(response)

if __name__ == '__main__':

    # Start the Flask server in a new thread
    threading.Thread(target=app.run, kwargs={"use_reloader": False,"debug":True,"host": "0.0.0.0"}).start()

小技巧

阿里云部署fastgpt oneapi,并且在本机映射autodl的ollama端口

• 后台运行

# 其中17001是本地端口  127.0.0.1:11434是autodl的端口
ssh -CNg -f -N  -L 17001:127.0.0.1:11434 [email protected] -p 35040

docker 运行m3e

docker run -d --name m3e -p 17002:6008 --gpus all -e sk-key=123321  registry.cn-hangzhou.aliyuncs.com/fastgpt_docker/m3e-large-api

错误解决

1. docker-compose up -d后oneapi不能启动

oneapi依赖mysql，主要是mysql的问题，另外启动一个mysql容器，oneapi依赖后就解决了。

2. oneapi配置大模型后，测试报错

{
    "message": "do request failed: Post \"http://localhost:11434/api/chat\": dial tcp [::1]:11434: connect: connection refused",
    "success": false,
    "time": 0
}

这是由于oneapi依赖mysql，配置的时候的填写本机真实ip地址，不能localhost或者127.0.0.1，正确配置如图