exclusive-qa-service.yml

ROSTemplateFormatVersion: '2015-09-01'
Description:
  en: DashVector and Tongyi QianWen build exclusive Q&A service.
  zh-cn: 向量检索与通义千问搭建专属问答服务。
Parameters:
  DashScopeApiKey:
    Type: String
    AssociationProperty: 'ALIYUN::DashScope::ApiKey'
    Label:
      en: DashScope API-KEY
      zh-cn: DashScope API-KEY
    Description:
      zh-cn: >-
        开通灵积模型服务，并获得 API-KEY。请参考：
        <a href="https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key" 
        target="_blank">开通DashScope并创建API-KEY</a>。
      en: >-
        Activate DashScope and obtain the API-KEY. Please refer to: 
        <a href="https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key" 
        target="_blank">Activate DashScope and create API-KEY</a>.
  DashVectorApiKey:
    Type: String
    AssociationProperty: 'ALIYUN::DashVector::ApiKey'
    Label:
      en: DashVector API-KEY
      zh-cn: DashVector API-KEY
    Description:
      zh-cn: >-
        开通DashVector向量检索服务，并获得 API-KEY。请参考：
        <a href="https://help.aliyun.com/document_detail/2510230.html" 
        target="_blank">开通DashVector并创建API-KEY</a>。
      en: >-
        Activate DashVector and obtain the API-KEY. Please refer to: 
        <a href="https://help.aliyun.com/document_detail/2510230.html" 
        target="_blank">Activate DashVector and create API-KEY</a>.
  ClusterName:
    Type: String
    Label:
      en: Name of DashVector Cluster
      zh-cn: DashVector集群名称
    AllowedPattern: '[-a-z0-9_]{3,32}'
    AssociationProperty: AutoCompleteInput
    AssociationPropertyMetadata:
      Length: 5
      Prefix: exclusive-qa-server-
      CharacterClasses:
        - Class: lowercase
          min: 1
    ConstraintDescription:
      zh-cn: 由大小写字母、数字、下划线(_)、中划线(-)组成，长度[3, 32]。
      en: It consists of uppercase and lowercase letters, numbers, underline (_), and dash (-). The length is [3, 32]
    Description:
      zh-cn: 本方案会创建一个免费版的集群，如果您账号下已经创建好了免费版的集群，请输入免费版集群的名称。
      en: >- 
        This solution will create a free version cluster. If you have already created 
        a free version cluster under your account, please enter the name of the free version cluster.
  ZoneId:
    Type: String
    AssociationProperty: 'ALIYUN::ECS::Instance::ZoneId'
    AssociationPropertyMetadata:
      AutoSelectFirst: true
    Label:
      en: Availability Zone
      zh-cn: 可用区
  InstanceType:
    Type: String
    AssociationProperty: 'ALIYUN::ECS::Instance::InstanceType'
    AssociationPropertyMetadata:
      SpotStrategy: SpotAsPriceGo
      InstanceChargeType: PostPaid
      SystemDiskCategory: cloud_essd
      ZoneId: ${ZoneId}
    Label:
      en: Instance Type
      zh-cn: 实例规格
    Description:
      zh-cn: 本方案会创建一个抢占式实例，并且自动部署专属问答服务。
      en: >-
        This solution will create a spot instance and automatically deploy a 
        exclusive question and answer service.
    Default: ecs.c7.large
  InstancePassword:
    NoEcho: true
    Type: String
    Description:
      en: >-
        Server login password, Length 8-30, must contain three(Capital letters,
        lowercase letters, numbers, ()`~!@#$%^&*_-+=|{}[]:;'<>,.?/ Special
        symbol in)
      zh-cn: >-
        服务器登录密码,长度8-30，必须包含三项（大写字母、小写字母、数字、 ()`~!@#$%^&*_-+=|{}[]:;'<>,.?/
        中的特殊符号）
    Label:
      en: Instance Password
      zh-cn: 实例密码
    ConstraintDescription:
      en: >-
        Length 8-30, must contain three(Capital letters, lowercase letters,
        numbers, ()`~!@#$%^&*_-+=|{}[]:;'<>,.?/ Special symbol in)
      zh-cn: '长度8-30，必须包含三项（大写字母、小写字母、数字、 ()`~!@#$%^&*_-+=|{}[]:;''<>,.?/ 中的特殊符号）'
    AssociationProperty: 'ALIYUN::ECS::Instance::Password'
    Default: null
Resources:
  Vpc:
    Type: 'ALIYUN::ECS::VPC'
    Properties:
      CidrBlock: 192.168.0.0/16
  VSwitch:
    Type: 'ALIYUN::ECS::VSwitch'
    Properties:
      VpcId:
        Ref: Vpc
      CidrBlock: 192.168.0.0/24
      ZoneId:
        Ref: ZoneId
  SecurityGroup:
    Type: 'ALIYUN::ECS::SecurityGroup'
    Properties:
      VpcId:
        Ref: Vpc
      SecurityGroupIngress:
        - PortRange: 22/22
          SourceCidrIp: 0.0.0.0/0
          IpProtocol: tcp
        - PortRange: 5000/5000
          SourceCidrIp: 0.0.0.0/0
          IpProtocol: tcp
  EcsInstance:
    Type: 'ALIYUN::ECS::InstanceGroup'
    Properties:
      VpcId:
        Ref: Vpc
      ZoneId:
        Ref: ZoneId
      VSwitchId:
        Ref: VSwitch
      SecurityGroupId:
        Ref: SecurityGroup
      ImageId: aliyun_3_9_x64_20G_alibase_
      InstanceName: dashVector
      InstanceType:
        Ref: InstanceType
      SystemDiskCategory: cloud_essd
      MaxAmount: 1
      InternetMaxBandwidthOut: 100
      SpotStrategy: SpotAsPriceGo
      Password:
        Ref: InstancePassword
  DashVectorCluster:
    Type: 'ALIYUN::DashVector::Cluster'
    Properties:
      IgnoreExisting: true
      ReplicaCount: 1
      ClusterName:
        Ref: ClusterName
      ClusterType: Serverless
  RunCommand:
    Type: 'ALIYUN::ECS::RunCommand'
    Properties:
      InstanceIds:
        Fn::GetAtt:
          - EcsInstance
          - InstanceIds
      Type: RunShellScript
      Sync: true
      Timeout: 3600
      CommandContent:
        Fn::Sub: |-
          #!/bin/bash
          
          # script exit code:
          # 0 - success
          # 1 - unsupported system
          # 2 - network not available
          # 3 - failed to git clone
          # 4 - failed to init python environment
          # 5 - failed to embedding data
          # 6 - failed to init git
          # 7 - failed to run flask app
          
          # 环境变量配置
          export PATH=/usr/local/bin:$PATH
          export DASHSCOPE_API_KEY=${DashScopeApiKey}
          export DASHVECTOR_API_KEY=${DashVectorApiKey}
          export DASHVECTOR_CLUSTER_ENDPOINT=${DashVectorCluster.Endpoint}
          
          # GitHub 仓库 URL
          REPO_URL="https://github.com/shijiebei2009/CEC-Corpus.git"
          # GitRepoName
          REPO_NAME="CEC-Corpus"
          # 最大尝试次数
          MAX_RETRIES=10
          # 当前尝试次数
          ATTEMPT_NUM=1
          # 等待时间 (秒) 在重试之前
          SLEEP_SECONDS=10
          # CollectionName
          COLLECTION_NAME="news_embeddings"
          
          function unsupported_system() {
              log_fatal 1 "Unsupported System: $1"
          }
          
          function log_info() {
              printf "%s [INFO] %s\n" "$(date '+%Y-%m-%d %H:%M:%S')" "$1"
          }
          
          function log_error() {
              printf "%s [ERROR] %s\n" "$(date '+%Y-%m-%d %H:%M:%S')" "$1"
          }
          
          function log_fatal() {
              printf "\n========================================================================\n"
              printf "%s [FATAL] %s\n" "$(date '+%Y-%m-%d %H:%M:%S')" "$2"
              printf "\n========================================================================\n"
              exit $1
          }
          
          function debug_exec(){
              local cmd="$@"
              log_info "$cmd"
              eval "$cmd"
              ret=$?
              echo ""
              log_info "$cmd, exit code: $ret"
              return $ret
          }
          
          function check_network_available() {
              log_info "ping github.com ..."
              if ! debug_exec ping -c 4 github.com; then
                  log_fatal 2 "Could not connect to https://github.com"
              fi
          }
          
          function clone_repo() {
              log_info "init git"
              if ! debug_exec yum install -y git && git init; then
                  log_fatal 6 "failed to init git"
              fi
          
              rm -rf $REPO_NAME
              until debug_exec git clone $REPO_URL; do
                  if [ "$ATTEMPT_NUM" -ge "$MAX_RETRIES" ]; then
                      log_fatal 3 "Reached maximum number of retries, aborting."
                  fi
                  log_error "Attempt $ATTEMPT_NUM failed! Trying again in $SLEEP_SECONDS seconds..."
                  sleep "$SLEEP_SECONDS"
                  ATTEMPT_NUM=$((ATTEMPT_NUM+1))
              done
          }
          
          function init_env() {
              log_info "init python environment"
              yum install -y python3.8
              ln -sf /usr/bin/python3.8 /usr/bin/python
              python -m ensurepip
              python -m pip install --upgrade pip
              python -m pip install -r requirements.txt
          }
          
          function embedding() {
              log_info "embedding data"
              if ! debug_exec python embedding.py; then
                  log_fatal 5 "failed to embedding data."
              fi
          }
          
          log_info "System Information:"
          if ! lsb_release -a; then
              unsupported_system
          fi;
          echo ""
          
          check_network_available
          
          cat << 'EOF' > requirements.txt
          flask==3.0.2
          PySocks==1.7.1
          dashscope==1.17.0
          dashvector==1.0.14
          
          EOF
          
          if ! debug_exec init_env; then
              log_fatal 4 "failed to init python environment"
          fi
          
          cat << EOF > embedding.py
          # -*- coding: utf-8 -*-
          
          import os
          import dashscope
          from dashscope import TextEmbedding
          from dashvector import Client, Doc
          
          
          def prepare_data(path, batch_size=25):
              batch_docs = []
              for file in os.listdir(path):
                  with open(path + '/' + file, 'r', encoding='utf-8') as f:
                      batch_docs.append(f.read())
                      if len(batch_docs) == batch_size:
                          yield batch_docs
                          batch_docs = []
          
              if batch_docs:
                  yield batch_docs
          
          
          def generate_embeddings(news):
              rsp = TextEmbedding.call(
                  model=TextEmbedding.Models.text_embedding_v1,
                  input=news
              )
              embeddings = [record['embedding'] for record in rsp.output['embeddings']]
              return embeddings if isinstance(news, list) else embeddings[0]
          
          
          if __name__ == '__main__':
              dashscope.api_key = os.environ.get('DASHSCOPE_API_KEY')
              collection_name = '$COLLECTION_NAME'
              # 初始化 dashvector client
              client = Client(
                  api_key=os.environ.get('DASHVECTOR_API_KEY'),
                  endpoint=os.environ.get('DASHVECTOR_CLUSTER_ENDPOINT')
              )
          
              try:
                  del_rsp = client.delete(collection_name)
                  assert del_rsp, del_rsp
              except AssertionError as e:
                  if e.args[0].code != -2021:
                      raise
          
              # 创建集合：指定集合名称和向量维度, text_embedding_v1 模型产生的向量统一为 1536 维
              rsp = client.create(collection_name, 1536)
              assert rsp, rsp
          
              # 加载语料
              id = 0
              collection = client.get(collection_name)
              for news in list(prepare_data('CEC-Corpus/raw corpus/allSourceText')):
                  ids = [id + i for i, _ in enumerate(news)]
                  id += len(news)
          
                  vectors = generate_embeddings(news)
                  # 写入 dashvector 构建索引
                  rsp = collection.upsert(
                      [
                          Doc(id=str(id), vector=vector, fields={"raw": doc})
                          for id, vector, doc in zip(ids, vectors, news)
                      ]
                  )
                  assert rsp, rsp
          
          EOF
          
          
          cat << EOF > run.py
          # -*- coding: utf-8 -*-
          
          import os
          import uuid
          from flask import Flask, request, render_template_string, redirect, url_for, session
          import dashscope
          from dashscope import TextEmbedding, Generation
          from dashvector import Client
          
          app = Flask(__name__)
          app.secret_key = uuid.uuid4().hex
          
          # HTML 模板
          HTML_TEMPLATE = '''
          <!doctype html>
          <html lang="en">
            <head>
              <title>DashVector</title>
              <style>
                body {
                  font-family: Arial, sans-serif;
                  margin: 0;
                  padding: 0;
                  background: #f7f7f7;
                }
                .container {
                  margin: auto;
                  padding: 20px;
                  max-width: 600px;
                  background: white;
                  box-shadow: 0 0 10px rgba(0, 0, 0, 0.1);
                }
                h1 {
                  color: #333;
                }
                input[type="text"], textarea {
                  width: 100%;
                  padding: 10px;
                  margin: 10px 0;
                  border: 1px solid #ddd;
                  border-radius: 5px;
                  box-sizing: border-box; /* 加上这个属性，防止 padding 喧嚣计算宽度 */
                }
                input[type="submit"] {
                  background: #5cb85c;
                  color: white;
                  padding: 10px 20px;
                  margin: 10px 0;
                  border: none;
                  border-radius: 5px;
                  cursor: pointer;
                }
                input[type="submit"]:hover {
                  background: #4cae4c;
                }
                .loader {
                  border: 5px solid #f3f3f3; /* Light grey */
                  border-top: 5px solid #3498db; /* Blue */
                  border-radius: 50%;
                  width: 50px;
                  height: 50px;
                  animation: spin 2s linear infinite;
                  margin: auto;
                  display: none;
                }
          
                @keyframes spin {
                  0% { transform: rotate(0deg); }
                  100% { transform: rotate(360deg); }
                }
              </style>
            </head>
            <body>
              <div class="container">
                <h1>基于专属知识的问答服务</h1>
                <form method="post" id="textForm">
                  <label for="inputText">请输入问题：</label>
                  <input type="text" id="inputText" name="inputText" placeholder="Enter..." value="{{ input }}">
                  <input type="submit" value="Submit">
                  <!-- 加载动画 -->
                  <div class="loader" id="loader"></div>
                </form>
                {% if output %}
                <h2>回答：</h2>
                <textarea id="outputText" name="outputText" rows="4" disabled>{{ output }}</textarea>
                {% endif %}
              </div>
          
              <script>
                // 使用 JavaScript 在提交表单时显示加载动画
                document.getElementById('textForm').onsubmit = function() {
                  document.getElementById('loader').style.display = 'block';
                };
              </script>
            </body>
          </html>
          '''
          
          
          @app.route('/', methods=['GET', 'POST'])
          def home():
              output_text = None
              input_text = "海南安定追尾事故，发生在哪里？原因是什么？人员伤亡情况如何"
              if 'output' in session:
                  output_text = session.pop('output', None)
                  input_text = session.get('input', None)
          
              if request.method == 'POST':
                  input_text = request.form.get('inputText')
                  output_text = get_answers(input_text)
                  session['output'] = output_text
                  session['input'] = input_text
                  return redirect(url_for('home'))
          
              return render_template_string(HTML_TEMPLATE, output=output_text, input=input_text)
          
          
          def get_answers(input_text):
              dashscope.api_key = os.environ.get('DASHSCOPE_API_KEY')
              # question = '海南安定追尾事故，发生在哪里？原因是什么？人员伤亡情况如何？'
              context = search_relevant_news(input_text)
              return answer_question(input_text, context)
          
          
          def search_relevant_news(question):
              # 初始化 dashvector client
              client = Client(
                api_key=os.environ.get('DASHVECTOR_API_KEY'),
                endpoint=os.environ.get('DASHVECTOR_CLUSTER_ENDPOINT')
              )
          
              # 获取刚刚存入的集合
              collection = client.get('$COLLECTION_NAME')
              assert collection, collection
          
              # 向量检索：指定 topk = 1
              rsp = collection.query(generate_embeddings(question), output_fields=['raw'], topk=1)
              assert rsp, rsp
              return rsp.output[0].fields['raw']
          
          
          def generate_embeddings(news):
              rsp = TextEmbedding.call(
                  model=TextEmbedding.Models.text_embedding_v1,
                  input=news
              )
              embeddings = [record['embedding'] for record in rsp.output['embeddings']]
              return embeddings if isinstance(news, list) else embeddings[0]
          
          
          def answer_question(question, context):
              prompt = f'''请基于\`\`\`内的内容回答问题。"
              \`\`\`
              {context}
              \`\`\`
              我的问题是：{question}。
              '''
          
              rsp = Generation.call(model='qwen-turbo', prompt=prompt)
              return rsp.output.text
          
          
          if __name__ == '__main__':
              app.run(host='0.0.0.0', debug=True)
          
          EOF
          
          clone_repo
          embedding
          
          if ! debug_exec "nohup python run.py > output.log 2>&1 &"; then
              log_fatal  "failed to run flask app"
          fi
Outputs:
  WebUrl:
    Description:
      zh-cn: Web 访问地址。
      en: The Addresses of Web.
    Value:
      'Fn::Sub':
        - 'http://${ServerAddress}:5000'
        - ServerAddress:
            'Fn::Select':
              - 0
              - 'Fn::GetAtt':
                  - EcsInstance
                  - PublicIps
  EcsLoginAddress:
    Description:
      zh-cn: ECS登录地址。
      en: Ecs login address.
    Value:
      'Fn::Sub':
        - >-
          https://ecs-workbench.aliyun.com/?from=EcsConsole&instanceType=ecs&regionId=${Region}&instanceId=${InstanceId}
        - InstanceId:
            'Fn::Select':
              - 0
              - 'Fn::GetAtt':
                  - EcsInstance
                  - InstanceIds
          Region:
            Ref: 'ALIYUN::Region'
Metadata:
  'ALIYUN::ROS::Interface':
    ParameterGroups:
      - Parameters:
          - DashScopeApiKey
          - DashVectorApiKey
          - ClusterName
        Label:
          default:
            en: ApiKey
            zh-cn: ApiKey配置
      - Parameters:
          - ZoneId
          - InstanceType
          - InstancePassword
        Label:
          default:
            en: Ecs Instance
            zh-cn: ECS实例配置
    TemplateTags:
      - acs:technical-solution:ai:向量检索与通义千问搭建专属问答服务-tech_solu_109