使用 GKE 和 Cloud Storage 构建 RAG 聊天机器人

本教程介绍如何将基于检索增强生成 (RAG) 的大语言模型 (LLM) 应用与上传到 Cloud Storage 存储桶的 PDF 文件集成。

本指南使用数据库作为存储和语义搜索引擎，用于保存所上传文档的表示法（嵌入）。您将使用 Langchain 框架与嵌入进行交互，并使用通过 Vertex AI 提供的 Gemini 模型。

Langchain 是一种流行的开源 Python 框架，可简化许多机器学习任务，并具有与不同矢量数据库和 AI 服务集成的界面。

本教程适用于想要将 RAG LLM 应用部署到 GKE 和 Cloud Storage 的云平台管理员和架构师、机器学习工程师以及 MLOps (DevOps) 专业人员。

创建集群

创建 Qdrant、Elasticsearch 或 Postgres 集群：

Qdrant

按照在 GKE 上部署 Qdrant 矢量数据库中的说明，创建在 Autopilot 模式或 Standard 模式 GKE 集群上运行的 Qdrant 集群。

Elasticsearch

按照在 GKE 上部署 Elasticsearch 矢量数据库中的说明，创建在 Autopilot 模式或 Standard 模式 GKE 集群上运行的 Elasticsearch 集群。

PGVector

按照在 GKE 上部署 PostgreSQL 矢量数据库中的说明，创建在 Autopilot 模式或 Standard 模式 GKE 集群上运行且包含 PGVector 的 Postgres 集群。

Weaviate

按照在 GKE 上部署 Weaviate 矢量数据库中的说明，创建在 Autopilot 或 Standard 模式 GKE 集群上运行的 Weaviate 集群。

设置环境

使用 Cloud Shell 设置您的环境：

  1. 设置项目的环境变量：

    Qdrant

    export PROJECT_ID=PROJECT_ID
export KUBERNETES_CLUSTER_PREFIX=qdrant
export CONTROL_PLANE_LOCATION=us-central1
export REGION=us-central1
export DB_NAMESPACE=qdrant

    将 PROJECT_ID 替换为您的Google Cloud 项目 ID。

    Elasticsearch 

    export PROJECT_ID=PROJECT_ID
export KUBERNETES_CLUSTER_PREFIX=elasticsearch
export CONTROL_PLANE_LOCATION=us-central1
export REGION=us-central1
export DB_NAMESPACE=elastic

    将 PROJECT_ID 替换为您的Google Cloud 项目 ID。

    PGVector 

    export PROJECT_ID=PROJECT_ID
export KUBERNETES_CLUSTER_PREFIX=postgres
export CONTROL_PLANE_LOCATION=us-central1
export REGION=us-central1
export DB_NAMESPACE=pg-ns

    将 PROJECT_ID 替换为您的Google Cloud 项目 ID。

    Weaviate

    export PROJECT_ID=PROJECT_ID
export KUBERNETES_CLUSTER_PREFIX=weaviate
export CONTROL_PLANE_LOCATION=us-central1
export REGION=us-central1
export DB_NAMESPACE=weaviate

    将 PROJECT_ID 替换为您的Google Cloud 项目 ID。

  2. 验证 GKE 集群是否正在运行：

    gcloud container clusters list --project=${PROJECT_ID} --location=${CONTROL_PLANE_LOCATION}

    输出类似于以下内容：

    NAME                                    LOCATION        MASTER_VERSION      MASTER_IP     MACHINE_TYPE  NODE_VERSION        NUM_NODES STATUS
[KUBERNETES_CLUSTER_PREFIX]-cluster   us-central1   1.30.1-gke.1329003  <EXTERNAL IP> e2-standard-2 1.30.1-gke.1329003   6        RUNNING

  3. 从 GitHub 克隆示例代码库：

    git clone https://github.com/GoogleCloudPlatform/kubernetes-engine-samples

  4. 导航到 databases 目录：

    cd kubernetes-engine-samples/databases

准备基础架构

创建 Artifact Registry 代码库，构建 Docker 映像，并将 Docker 映像推送到 Artifact Registry：

  1. 创建 Artifact Registry 代码库：

    gcloud artifacts repositories create ${KUBERNETES_CLUSTER_PREFIX}-images \
    --repository-format=docker \
    --location=${REGION} \
    --description="Vector database images repository" \
    --async

  2. 为 Compute Engine 服务账号设置 storage.objectAdmin 和 artifactregistry.admin 权限，以使用 Cloud Build 为 embed-docs 和 chatbot Service 构建和推送 Docker 映像。

    export PROJECT_NUMBER=PROJECT_NUMBER

gcloud projects add-iam-policy-binding ${PROJECT_ID}  \
--member="serviceAccount:${PROJECT_NUMBER}-compute@developer.gserviceaccount.com" \
--role="roles/storage.objectAdmin"

gcloud projects add-iam-policy-binding ${PROJECT_ID}  \
--member="serviceAccount:${PROJECT_NUMBER}-compute@developer.gserviceaccount.com" \
--role="roles/artifactregistry.admin"

    将 PROJECT_NUMBER 替换为您的Google Cloud 项目编号。

  3. 为 embed-docs 和 chatbot Service 构建 Docker 映像。 embed-docs 映像包含接收 Eventarc 转发器请求的应用和嵌入作业的 Python 代码。

    Qdrant

    export DOCKER_REPO="${REGION}-docker.pkg.dev/${PROJECT_ID}/${KUBERNETES_CLUSTER_PREFIX}-images"
gcloud builds submit qdrant/docker/chatbot --region=${REGION} \
  --tag ${DOCKER_REPO}/chatbot:1.0 --async
gcloud builds submit qdrant/docker/embed-docs --region=${REGION} \
  --tag ${DOCKER_REPO}/embed-docs:1.0 --async

    Elasticsearch 

    export DOCKER_REPO="${REGION}-docker.pkg.dev/${PROJECT_ID}/${KUBERNETES_CLUSTER_PREFIX}-images"
gcloud builds submit elasticsearch/docker/chatbot --region=${REGION} \
  --tag ${DOCKER_REPO}/chatbot:1.0 --async
gcloud builds submit elasticsearch/docker/embed-docs --region=${REGION} \
  --tag ${DOCKER_REPO}/embed-docs:1.0 --async

    PGVector

    export DOCKER_REPO="${REGION}-docker.pkg.dev/${PROJECT_ID}/${KUBERNETES_CLUSTER_PREFIX}-images"
gcloud builds submit postgres-pgvector/docker/chatbot --region=${REGION} \
  --tag ${DOCKER_REPO}/chatbot:1.0 --async
gcloud builds submit postgres-pgvector/docker/embed-docs --region=${REGION} \
  --tag ${DOCKER_REPO}/embed-docs:1.0 --async

    Weaviate

    export DOCKER_REPO="${REGION}-docker.pkg.dev/${PROJECT_ID}/${KUBERNETES_CLUSTER_PREFIX}-images"
gcloud builds submit weaviate/docker/chatbot --region=${REGION} \
  --tag ${DOCKER_REPO}/chatbot:1.0 --async
gcloud builds submit weaviate/docker/embed-docs --region=${REGION} \
  --tag ${DOCKER_REPO}/embed-docs:1.0 --async

  4. 验证映像：

    gcloud artifacts docker images list $DOCKER_REPO \
    --project=$PROJECT_ID \
    --format="value(IMAGE)"

    输出类似于以下内容：

    $REGION-docker.pkg.dev/$PROJECT_ID/${KUBERNETES_CLUSTER_PREFIX}-images/chatbot
$REGION-docker.pkg.dev/$PROJECT_ID/${KUBERNETES_CLUSTER_PREFIX}-images/embed-docs

  5. 部署具有运行 Kubernetes 作业的权限的 Kubernetes 服务账号：

    Qdrant

    sed "s/<PROJECT_ID>/$PROJECT_ID/;s/<CLUSTER_PREFIX>/$KUBERNETES_CLUSTER_PREFIX/" qdrant/manifests/05-rag/service-account.yaml | kubectl -n qdrant apply -f -

    Elasticsearch 

    sed "s/<PROJECT_ID>/$PROJECT_ID/;s/<CLUSTER_PREFIX>/$KUBERNETES_CLUSTER_PREFIX/" elasticsearch/manifests/05-rag/service-account.yaml | kubectl -n elastic apply -f -

    PGVector

    sed "s/<PROJECT_ID>/$PROJECT_ID/;s/<CLUSTER_PREFIX>/$KUBERNETES_CLUSTER_PREFIX/" postgres-pgvector/manifests/03-rag/service-account.yaml | kubectl -n pg-ns apply -f -

    Weaviate

    sed "s/<PROJECT_ID>/$PROJECT_ID/;s/<CLUSTER_PREFIX>/$KUBERNETES_CLUSTER_PREFIX/" weaviate/manifests/04-rag/service-account.yaml | kubectl -n weaviate apply -f -

  6. 使用 Terraform 创建 GKE 集群并将 create_service_account 设置为 true 时，系统会创建一个单独的服务账号，供集群和节点使用。向此 Compute Engine 服务账号授予 artifactregistry.serviceAgent 角色，以允许节点从为 embed-docs 和 chatbot 创建的 Artifact Registry 中拉取映像。

    export CLUSTER_SERVICE_ACCOUNT=$(gcloud container clusters describe ${KUBERNETES_CLUSTER_PREFIX}-cluster \
--location=${CONTROL_PLANE_LOCATION} \
--format="value(nodeConfig.serviceAccount)")

gcloud projects add-iam-policy-binding ${PROJECT_ID}  \
--member="serviceAccount:${CLUSTER_SERVICE_ACCOUNT}" \
--role="roles/artifactregistry.serviceAgent"

    如果未向该服务账号授予访问权限，则部署 embed-docs 和 chatbot Service 时，节点在尝试从 Artifact Registry 拉取映像时可能会遇到权限问题。

  7. 为 embed-docs 和 chatbot Service 部署 Kubernetes Deployment。Deployment 是一个 Kubernetes API 对象，可让您运行在集群节点中分布的多个 Pod 副本。

    Qdrant 

    sed "s|<DOCKER_REPO>|$DOCKER_REPO|" qdrant/manifests/05-rag/chatbot.yaml | kubectl -n qdrant apply -f -
sed "s|<DOCKER_REPO>|$DOCKER_REPO|" qdrant/manifests/05-rag/docs-embedder.yaml | kubectl -n qdrant apply -f -

    Elasticsearch 

    sed "s|<DOCKER_REPO>|$DOCKER_REPO|" elasticsearch/manifests/05-rag/chatbot.yaml | kubectl -n elastic apply -f -
sed "s|<DOCKER_REPO>|$DOCKER_REPO|" elasticsearch/manifests/05-rag/docs-embedder.yaml | kubectl -n elastic apply -f -

    PGVector

    sed "s|<DOCKER_REPO>|$DOCKER_REPO|" postgres-pgvector/manifests/03-rag/chatbot.yaml | kubectl -n pg-ns apply -f -
sed "s|<DOCKER_REPO>|$DOCKER_REPO|" postgres-pgvector/manifests/03-rag/docs-embedder.yaml | kubectl -n pg-ns apply -f -

    Weaviate

    sed "s|<DOCKER_REPO>|$DOCKER_REPO|" weaviate/manifests/04-rag/chatbot.yaml | kubectl -n weaviate apply -f -
sed "s|<DOCKER_REPO>|$DOCKER_REPO|" weaviate/manifests/04-rag/docs-embedder.yaml | kubectl -n weaviate apply -f -

  8. 为 GKE 启用 Eventarc 触发器：

    gcloud eventarc gke-destinations init

    出现提示时，输入 y。

  9. 使用 Terraform 部署 Cloud Storage 存储桶并创建 Eventarc 触发器：

    export GOOGLE_OAUTH_ACCESS_TOKEN=$(gcloud auth print-access-token)
terraform -chdir=vector-database/terraform/cloud-storage init
terraform -chdir=vector-database/terraform/cloud-storage apply \
  -var project_id=${PROJECT_ID} \
  -var region=${REGION} \
  -var cluster_prefix=${KUBERNETES_CLUSTER_PREFIX} \
  -var db_namespace=${DB_NAMESPACE}

    出现提示时，请输入 yes。该命令可能需要几分钟才能完成。

    Terraform 会创建以下资源：

    • 用于上传文档的 Cloud Storage 存储桶
    • Eventarc 触发器
    • 名为 service_account_eventarc_name 且有权使用 Eventarc 的 Google Cloud 服务账号。
    • 名为 service_account_bucket_name 且有权读取存储桶和访问 Vertex AI 模型的 Google Cloud 服务账号。

    输出内容类似如下：

    ... # Several lines of output omitted

Apply complete! Resources: 15 added, 0 changed, 0 destroyed.

... # Several lines of output omitted

加载文档并运行聊天机器人查询

上传演示文档，然后运行查询以使用聊天机器人搜索演示文档：

  1. 将示例 carbon-free-energy.pdf 文档上传到您的存储桶：

    gcloud storage cp vector-database/documents/carbon-free-energy.pdf gs://${PROJECT_ID}-${KUBERNETES_CLUSTER_PREFIX}-training-docs

  2. 验证文档嵌入器作业是否已成功完成：

    kubectl get job -n ${DB_NAMESPACE}

    输出类似于以下内容：

    NAME                            COMPLETIONS   DURATION   AGE
docs-embedder1716570453361446   1/1           32s        71s

  3. 获取负载均衡器的外部 IP 地址：

    export EXTERNAL_IP=$(kubectl -n ${DB_NAMESPACE} get svc chatbot --output jsonpath='{.status.loadBalancer.ingress[0].ip}')
echo http://${EXTERNAL_IP}:80

  4. 在网络浏览器中打开外部 IP 地址：

    http://EXTERNAL_IP

    聊天机器人会返回类似于如下所示的消息：

    How can I help you?

  5. 提出与所上传文档的内容相关的问题。如果聊天机器人找不到任何内容，则会回答 I don't know。例如，您可能会提出以下问题：

    You: Hi, what are Google plans for the future?

    聊天机器人的示例输出类似于以下内容：

    Bot: Google intends to run on carbon-free energy everywhere, at all times by 2030. To achieve this, it will rely on a combination of renewable energy sources, such as wind and solar, and carbon-free technologies, such as battery storage.

  6. 向聊天机器人提出与上传的文档无关的问题。例如，您可能会提出以下问题：

    You: What are Google plans to colonize Mars?

    聊天机器人的示例输出类似于以下内容：

    Bot: I don't know. The provided context does not mention anything about Google's plans to colonize Mars.

关于应用代码

本部分介绍应用代码的工作原理。Docker 映像中包含三个脚本：

  • endpoint.py：在每次上传文档时接收 Eventarc 事件，并启动 Kubernetes 作业来处理这些事件。
  • embedding-job.py：从存储桶下载文档，创建嵌入，然后将嵌入插入到矢量数据库中。
  • chat.py：对存储的文档内容运行查询。

该图展示了使用文档数据生成回答的过程：

在该图中，应用会加载 PDF 文件，将文件拆分为多个分块，然后将这些分块转换为矢量，最后将相应矢量发送到矢量数据库。之后，用户向聊天机器人提问。RAG 链使用语义搜索来搜索矢量数据库，然后将上下文与问题一起返回给 LLM。LLM 会回答问题，并将问题存储到聊天记录中。

关于 endpoint.py

此文件处理来自 Eventarc 的消息，创建用于嵌入文档的 Kubernetes 作业，并通过端口 5001 接受来自任何位置的请求

Qdrant

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from flask import Flask, jsonify
from flask import request
import logging
import sys,os, time
from kubernetes import client, config, utils
import kubernetes.client
from kubernetes.client.rest import ApiException


app = Flask(__name__)
@app.route('/check')
def message():
    return jsonify({"Message": "Hi there"})


@app.route('/', methods=['POST'])
def bucket():
    request_data = request.get_json()
    print(request_data)
    bckt = request_data['bucket']
    f_name = request_data['name']
    id = request_data['generation'] 
    kube_create_job(bckt, f_name, id)
    return "ok"

# Set logging
logging.basicConfig(stream=sys.stdout, level=logging.INFO)

# Setup K8 configs
config.load_incluster_config()
def kube_create_job_object(name, container_image, bucket_name, f_name, namespace="qdrant", container_name="jobcontainer", env_vars={}):

    body = client.V1Job(api_version="batch/v1", kind="Job")
    body.metadata = client.V1ObjectMeta(namespace=namespace, name=name)
    body.status = client.V1JobStatus()

    template = client.V1PodTemplate()
    template.template = client.V1PodTemplateSpec()
    env_list = [
        client.V1EnvVar(name="QDRANT_URL", value=os.getenv("QDRANT_URL")),
        client.V1EnvVar(name="COLLECTION_NAME", value="training-docs"), 
        client.V1EnvVar(name="FILE_NAME", value=f_name), 
        client.V1EnvVar(name="BUCKET_NAME", value=bucket_name),
        client.V1EnvVar(name="APIKEY", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="api-key", name="qdrant-database-apikey"))), 
    ]

    container = client.V1Container(name=container_name, image=container_image, env=env_list)
    template.template.spec = client.V1PodSpec(containers=[container], restart_policy='Never', service_account='embed-docs-sa')

    body.spec = client.V1JobSpec(backoff_limit=3, ttl_seconds_after_finished=60, template=template.template)
    return body
def kube_test_credentials():
    try: 
        api_response = api_instance.get_api_resources()
        logging.info(api_response)
    except ApiException as e:
        print("Exception when calling API: %s\n" % e)

def kube_create_job(bckt, f_name, id):
    container_image = os.getenv("JOB_IMAGE")
    namespace = os.getenv("JOB_NAMESPACE")
    name = "docs-embedder" + id
    body = kube_create_job_object(name, container_image, bckt, f_name)
    v1=client.BatchV1Api()
    try: 
        v1.create_namespaced_job(namespace, body, pretty=True)
    except ApiException as e:
        print("Exception when calling BatchV1Api->create_namespaced_job: %s\n" % e)
    return

if __name__ == '__main__':
    app.run('0.0.0.0', port=5001, debug=True)

Elasticsearch 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from flask import Flask, jsonify
from flask import request
import logging
import sys,os, time
from kubernetes import client, config, utils
import kubernetes.client
from kubernetes.client.rest import ApiException


app = Flask(__name__)
@app.route('/check')
def message():
    return jsonify({"Message": "Hi there"})


@app.route('/', methods=['POST'])
def bucket():
    request_data = request.get_json()
    print(request_data)
    bckt = request_data['bucket']
    f_name = request_data['name']
    id = request_data['generation'] 
    kube_create_job(bckt, f_name, id)
    return "ok"

# Set logging
logging.basicConfig(stream=sys.stdout, level=logging.INFO)

# Setup K8 configs
config.load_incluster_config()

def kube_create_job_object(name, container_image, bucket_name, f_name, namespace="elastic", container_name="jobcontainer", env_vars={}):

    body = client.V1Job(api_version="batch/v1", kind="Job")
    body.metadata = client.V1ObjectMeta(namespace=namespace, name=name)
    body.status = client.V1JobStatus()

    template = client.V1PodTemplate()
    template.template = client.V1PodTemplateSpec()
    env_list = [
        client.V1EnvVar(name="ES_URL", value=os.getenv("ES_URL")),
        client.V1EnvVar(name="INDEX_NAME", value="training-docs"), 
        client.V1EnvVar(name="FILE_NAME", value=f_name), 
        client.V1EnvVar(name="BUCKET_NAME", value=bucket_name),
        client.V1EnvVar(name="PASSWORD", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="elastic", name="elasticsearch-ha-es-elastic-user"))), 
    ]

    container = client.V1Container(name=container_name, image=container_image, image_pull_policy='Always', env=env_list)
    template.template.spec = client.V1PodSpec(containers=[container], restart_policy='Never', service_account='embed-docs-sa')

    body.spec = client.V1JobSpec(backoff_limit=3, ttl_seconds_after_finished=60, template=template.template)
    return body

def kube_test_credentials():
    try: 
        api_response = api_instance.get_api_resources()
        logging.info(api_response)
    except ApiException as e:
        print("Exception when calling API: %s\n" % e)

def kube_create_job(bckt, f_name, id):
    container_image = os.getenv("JOB_IMAGE")
    namespace = os.getenv("JOB_NAMESPACE")
    name = "docs-embedder" + id
    body = kube_create_job_object(name, container_image, bckt, f_name)
    v1=client.BatchV1Api()
    try: 
        v1.create_namespaced_job(namespace, body, pretty=True)
    except ApiException as e:
        print("Exception when calling BatchV1Api->create_namespaced_job: %s\n" % e)
    return

if __name__ == '__main__':
    app.run('0.0.0.0', port=5001, debug=True)

PGVector 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from flask import Flask, jsonify
from flask import request
import logging
import sys,os, time
from kubernetes import client, config, utils
import kubernetes.client
from kubernetes.client.rest import ApiException


app = Flask(__name__)
@app.route('/check')
def message():
    return jsonify({"Message": "Hi there"})


@app.route('/', methods=['POST'])
def bucket():
    request_data = request.get_json()
    print(request_data)
    bckt = request_data['bucket']
    f_name = request_data['name']
    id = request_data['generation'] 
    kube_create_job(bckt, f_name, id)
    return "ok"

# Set logging
logging.basicConfig(stream=sys.stdout, level=logging.INFO)

# Setup K8 configs
config.load_incluster_config()
def kube_create_job_object(name, container_image, bucket_name, f_name, namespace="pg-ns", container_name="jobcontainer", env_vars={}):

    body = client.V1Job(api_version="batch/v1", kind="Job")
    body.metadata = client.V1ObjectMeta(namespace=namespace, name=name)
    body.status = client.V1JobStatus()

    template = client.V1PodTemplate()
    template.template = client.V1PodTemplateSpec()
    env_list = [
        client.V1EnvVar(name="POSTGRES_HOST", value=os.getenv("POSTGRES_HOST")),
        client.V1EnvVar(name="DATABASE_NAME", value="app"), 
        client.V1EnvVar(name="COLLECTION_NAME", value="training-docs"), 
        client.V1EnvVar(name="FILE_NAME", value=f_name), 
        client.V1EnvVar(name="BUCKET_NAME", value=bucket_name),
        client.V1EnvVar(name="PASSWORD", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="password", name="gke-pg-cluster-app"))), 
        client.V1EnvVar(name="USERNAME", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="username", name="gke-pg-cluster-app"))), 
    ]

    container = client.V1Container(name=container_name, image=container_image, image_pull_policy='Always', env=env_list)
    template.template.spec = client.V1PodSpec(containers=[container], restart_policy='Never', service_account='embed-docs-sa')

    body.spec = client.V1JobSpec(backoff_limit=3, ttl_seconds_after_finished=60, template=template.template)
    return body
def kube_test_credentials():
    try: 
        api_response = api_instance.get_api_resources()
        logging.info(api_response)
    except ApiException as e:
        print("Exception when calling API: %s\n" % e)

def kube_create_job(bckt, f_name, id):
    container_image = os.getenv("JOB_IMAGE")
    namespace = os.getenv("JOB_NAMESPACE")
    name = "docs-embedder" + id
    body = kube_create_job_object(name, container_image, bckt, f_name)
    v1=client.BatchV1Api()
    try: 
        v1.create_namespaced_job(namespace, body, pretty=True)
    except ApiException as e:
        print("Exception when calling BatchV1Api->create_namespaced_job: %s\n" % e)
    return

if __name__ == '__main__':
    app.run('0.0.0.0', port=5001, debug=True)

Weaviate 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from flask import Flask, jsonify
from flask import request
import logging
import sys,os, time
from kubernetes import client, config, utils
import kubernetes.client
from kubernetes.client.rest import ApiException


app = Flask(__name__)
@app.route('/check')
def message():
    return jsonify({"Message": "Hi there"})


@app.route('/', methods=['POST'])
def bucket():
    request_data = request.get_json()
    print(request_data)
    bckt = request_data['bucket']
    f_name = request_data['name']
    id = request_data['generation'] 
    kube_create_job(bckt, f_name, id)
    return "ok"

# Set logging
logging.basicConfig(stream=sys.stdout, level=logging.INFO)

# Setup K8 configs
config.load_incluster_config()
def kube_create_job_object(name, container_image, bucket_name, f_name, namespace, container_name="jobcontainer", env_vars={}):

    body = client.V1Job(api_version="batch/v1", kind="Job")
    body.metadata = client.V1ObjectMeta(namespace=namespace, name=name)
    body.status = client.V1JobStatus()

    template = client.V1PodTemplate()
    template.template = client.V1PodTemplateSpec()
    env_list = [
        client.V1EnvVar(name="WEAVIATE_ENDPOINT", value=os.getenv("WEAVIATE_ENDPOINT")),
        client.V1EnvVar(name="WEAVIATE_GRPC_ENDPOINT", value=os.getenv("WEAVIATE_GRPC_ENDPOINT")),
        client.V1EnvVar(name="FILE_NAME", value=f_name), 
        client.V1EnvVar(name="BUCKET_NAME", value=bucket_name),
        client.V1EnvVar(name="APIKEY", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="AUTHENTICATION_APIKEY_ALLOWED_KEYS", name="apikeys"))), 
    ]

    container = client.V1Container(name=container_name, image=container_image, image_pull_policy='Always', env=env_list)
    template.template.spec = client.V1PodSpec(containers=[container], restart_policy='Never', service_account='embed-docs-sa')

    body.spec = client.V1JobSpec(backoff_limit=3, ttl_seconds_after_finished=60, template=template.template)
    return body
def kube_test_credentials():
    try: 
        api_response = api_instance.get_api_resources()
        logging.info(api_response)
    except ApiException as e:
        print("Exception when calling API: %s\n" % e)

def kube_create_job(bckt, f_name, id):
    container_image = os.getenv("JOB_IMAGE")
    namespace = os.getenv("JOB_NAMESPACE")
    name = "docs-embedder" + id
    body = kube_create_job_object(name, container_image, bckt, f_name, namespace)
    v1=client.BatchV1Api()
    try: 
        v1.create_namespaced_job(namespace, body, pretty=True)
    except ApiException as e:
        print("Exception when calling BatchV1Api->create_namespaced_job: %s\n" % e)
    return

if __name__ == '__main__':
    app.run('0.0.0.0', port=5001, debug=True)

关于 embedding-job.py

此文件处理文档并将其发送到矢量数据库。

Qdrant

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import ChatVertexAI
from langchain.prompts import ChatPromptTemplate
from langchain_google_vertexai import VertexAIEmbeddings
from langchain.memory import ConversationBufferWindowMemory
from langchain_community.vectorstores import Qdrant
from qdrant_client import QdrantClient
import streamlit as st
import os

vertexAI = ChatVertexAI(model_name=os.getenv("VERTEX_AI_MODEL_NAME", "gemini-2.5-flash-preview-04-17"), streaming=True, convert_system_message_to_human=True)
prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant who helps in finding answers to questions using the provided context."),
        ("human", """
        The answer should be based on the text context given in "text_context" and the conversation history given in "conversation_history" along with its Caption: \n
        Base your response on the provided text context and the current conversation history to answer the query.
        Select the most relevant information from the context.
        Generate a draft response using the selected information. Remove duplicate content from the draft response.
        Generate your final response after adjusting it to increase accuracy and relevance.
        Now only show your final response!
        If you do not know the answer or context is not relevant, response with "I don't know".

        text_context:
        {context}

        conversation_history:
        {history}

        query:
        {query}
        """),
    ]
)

embedding_model = VertexAIEmbeddings("text-embedding-005")

client = QdrantClient(
    url=os.getenv("QDRANT_URL"),
    api_key=os.getenv("APIKEY"),
)
collection_name = os.getenv("COLLECTION_NAME")
vector_search = Qdrant(client, collection_name, embeddings=embedding_model)
def format_docs(docs):
    return "\n\n".join([d.page_content for d in docs])

st.title("🤖 Chatbot")
if "messages" not in st.session_state:
    st.session_state["messages"] = [{"role": "ai", "content": "How can I help you?"}]
if "memory" not in st.session_state:
    st.session_state["memory"] = ConversationBufferWindowMemory(
        memory_key="history",
        ai_prefix="Bob",
        human_prefix="User",
        k=3,
    )
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.write(message["content"])
if chat_input := st.chat_input():
    with st.chat_message("human"):
        st.write(chat_input)
        st.session_state.messages.append({"role": "human", "content": chat_input})

    found_docs = vector_search.similarity_search(chat_input)
    context = format_docs(found_docs)

    prompt_value = prompt_template.format_messages(name="Bob", query=chat_input, context=context, history=st.session_state.memory.load_memory_variables({}))
    with st.chat_message("ai"):
        with st.spinner("Typing..."):
            content = ""
            with st.empty():
                for chunk in vertexAI.stream(prompt_value):
                    content += chunk.content
                    st.write(content)
            st.session_state.messages.append({"role": "ai", "content": content})

    st.session_state.memory.save_context({"input": chat_input}, {"output": content})

Elasticsearch 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import VertexAIEmbeddings
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from elasticsearch import Elasticsearch
from langchain_community.vectorstores.elasticsearch import ElasticsearchStore
from google.cloud import storage
import os

bucketname = os.getenv("BUCKET_NAME")
filename = os.getenv("FILE_NAME")

storage_client = storage.Client()
bucket = storage_client.bucket(bucketname)
blob = bucket.blob(filename)
blob.download_to_filename("/documents/" + filename)

loader = PyPDFLoader("/documents/" + filename)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = loader.load_and_split(text_splitter)

embeddings = VertexAIEmbeddings("text-embedding-005")

client = Elasticsearch(
    [os.getenv("ES_URL")], 
    verify_certs=False, 
    ssl_show_warn=False,
    basic_auth=("elastic", os.getenv("PASSWORD"))
)

db = ElasticsearchStore.from_documents(
    documents,
    embeddings,
    es_connection=client,
    index_name=os.getenv("INDEX_NAME")
)
db.client.indices.refresh(index=os.getenv("INDEX_NAME"))

print(filename + " was successfully embedded") 
print(f"# of vectors = {len(documents)}")

PGVector 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import VertexAIEmbeddings
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores.pgvector import PGVector
from google.cloud import storage
import os
bucketname = os.getenv("BUCKET_NAME")
filename = os.getenv("FILE_NAME")

storage_client = storage.Client()
bucket = storage_client.bucket(bucketname)
blob = bucket.blob(filename)
blob.download_to_filename("/documents/" + filename)

loader = PyPDFLoader("/documents/" + filename)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = loader.load_and_split(text_splitter)
for document in documents:
    document.page_content = document.page_content.replace('\x00', '')

embeddings = VertexAIEmbeddings("text-embedding-005")

CONNECTION_STRING = PGVector.connection_string_from_db_params(
    driver="psycopg2",
    host=os.environ.get("POSTGRES_HOST"),
    port=5432,
    database=os.environ.get("DATABASE_NAME"),
    user=os.environ.get("USERNAME"),
    password=os.environ.get("PASSWORD"),
)
COLLECTION_NAME = os.environ.get("COLLECTION_NAME")

db = PGVector.from_documents(
    embedding=embeddings,
    documents=documents,
    collection_name=COLLECTION_NAME,
    connection_string=CONNECTION_STRING,
    use_jsonb=True
)

print(filename + " was successfully embedded") 
print(f"# of vectors = {len(documents)}")

Weaviate 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import VertexAIEmbeddings
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
import weaviate
from weaviate.connect import ConnectionParams
from langchain_weaviate.vectorstores import WeaviateVectorStore
from google.cloud import storage
import os
bucketname = os.getenv("BUCKET_NAME")
filename = os.getenv("FILE_NAME")

storage_client = storage.Client()
bucket = storage_client.bucket(bucketname)
blob = bucket.blob(filename)
blob.download_to_filename("/documents/" + filename)

loader = PyPDFLoader("/documents/" + filename)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = loader.load_and_split(text_splitter)

embeddings = VertexAIEmbeddings("text-embedding-005")

auth_config = weaviate.auth.AuthApiKey(api_key=os.getenv("APIKEY"))
client = weaviate.WeaviateClient(
    connection_params=ConnectionParams.from_params(
        http_host=os.getenv("WEAVIATE_ENDPOINT"),
        http_port="80",
        http_secure=False,
        grpc_host=os.getenv("WEAVIATE_GRPC_ENDPOINT"),
        grpc_port="50051",
        grpc_secure=False,
    ),
    auth_client_secret=auth_config
)
client.connect()
if not client.collections.exists("trainingdocs"):
    collection = client.collections.create(name="trainingdocs")
db = WeaviateVectorStore.from_documents(documents, embeddings, client=client, index_name="trainingdocs")

print(filename + " was successfully embedded") 
print(f"# of vectors = {len(documents)}")

关于 chat.py

此文件将模型配置为仅使用提供的上下文和以前的回答来回答问题。如果上下文或对话历史记录与任何数据都不匹配，则模型会返回 I don't know。

Qdrant

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from flask import Flask, jsonify
from flask import request
import logging
import sys,os, time
from kubernetes import client, config, utils
import kubernetes.client
from kubernetes.client.rest import ApiException


app = Flask(__name__)
@app.route('/check')
def message():
    return jsonify({"Message": "Hi there"})


@app.route('/', methods=['POST'])
def bucket():
    request_data = request.get_json()
    print(request_data)
    bckt = request_data['bucket']
    f_name = request_data['name']
    id = request_data['generation'] 
    kube_create_job(bckt, f_name, id)
    return "ok"

# Set logging
logging.basicConfig(stream=sys.stdout, level=logging.INFO)

# Setup K8 configs
config.load_incluster_config()
def kube_create_job_object(name, container_image, bucket_name, f_name, namespace="qdrant", container_name="jobcontainer", env_vars={}):

    body = client.V1Job(api_version="batch/v1", kind="Job")
    body.metadata = client.V1ObjectMeta(namespace=namespace, name=name)
    body.status = client.V1JobStatus()

    template = client.V1PodTemplate()
    template.template = client.V1PodTemplateSpec()
    env_list = [
        client.V1EnvVar(name="QDRANT_URL", value=os.getenv("QDRANT_URL")),
        client.V1EnvVar(name="COLLECTION_NAME", value="training-docs"), 
        client.V1EnvVar(name="FILE_NAME", value=f_name), 
        client.V1EnvVar(name="BUCKET_NAME", value=bucket_name),
        client.V1EnvVar(name="APIKEY", value_from=client.V1EnvVarSource(secret_key_ref=client.V1SecretKeySelector(key="api-key", name="qdrant-database-apikey"))), 
    ]

    container = client.V1Container(name=container_name, image=container_image, env=env_list)
    template.template.spec = client.V1PodSpec(containers=[container], restart_policy='Never', service_account='embed-docs-sa')

    body.spec = client.V1JobSpec(backoff_limit=3, ttl_seconds_after_finished=60, template=template.template)
    return body
def kube_test_credentials():
    try: 
        api_response = api_instance.get_api_resources()
        logging.info(api_response)
    except ApiException as e:
        print("Exception when calling API: %s\n" % e)

def kube_create_job(bckt, f_name, id):
    container_image = os.getenv("JOB_IMAGE")
    namespace = os.getenv("JOB_NAMESPACE")
    name = "docs-embedder" + id
    body = kube_create_job_object(name, container_image, bckt, f_name)
    v1=client.BatchV1Api()
    try: 
        v1.create_namespaced_job(namespace, body, pretty=True)
    except ApiException as e:
        print("Exception when calling BatchV1Api->create_namespaced_job: %s\n" % e)
    return

if __name__ == '__main__':
    app.run('0.0.0.0', port=5001, debug=True)

Elasticsearch 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import ChatVertexAI
from langchain.prompts import ChatPromptTemplate
from langchain_google_vertexai import VertexAIEmbeddings
from langchain.memory import ConversationBufferWindowMemory
from elasticsearch import Elasticsearch
from langchain_community.vectorstores.elasticsearch import ElasticsearchStore
import streamlit as st
import os

vertexAI = ChatVertexAI(model_name=os.getenv("VERTEX_AI_MODEL_NAME", "gemini-2.5-flash-preview-04-17"), streaming=True, convert_system_message_to_human=True)
prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant who helps in finding answers to questions using the provided context."),
        ("human", """
        The answer should be based on the text context given in "text_context" and the conversation history given in "conversation_history" along with its Caption: \n
        Base your response on the provided text context and the current conversation history to answer the query.
        Select the most relevant information from the context.
        Generate a draft response using the selected information. Remove duplicate content from the draft response.
        Generate your final response after adjusting it to increase accuracy and relevance.
        Now only show your final response!
        If you do not know the answer or context is not relevant, response with "I don't know".

        text_context:
        {context}

        conversation_history:
        {history}

        query:
        {query}
        """),
    ]
)

embedding_model = VertexAIEmbeddings("text-embedding-005")

client = Elasticsearch(
    [os.getenv("ES_URL")], 
    verify_certs=False, 
    ssl_show_warn=False,
    basic_auth=("elastic", os.getenv("PASSWORD"))
)
vector_search = ElasticsearchStore(
    index_name=os.getenv("INDEX_NAME"),
    es_connection=client,
    embedding=embedding_model
)

def format_docs(docs):
    return "\n\n".join([d.page_content for d in docs])

st.title("🤖 Chatbot")
if "messages" not in st.session_state:
    st.session_state["messages"] = [{"role": "ai", "content": "How can I help you?"}]

if "memory" not in st.session_state:
    st.session_state["memory"] = ConversationBufferWindowMemory(
        memory_key="history",
        ai_prefix="Bot",
        human_prefix="User",
        k=3,
    )

for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.write(message["content"])

if chat_input := st.chat_input():
    with st.chat_message("human"):
        st.write(chat_input)
        st.session_state.messages.append({"role": "human", "content": chat_input})

    found_docs = vector_search.similarity_search(chat_input)
    context = format_docs(found_docs)

    prompt_value = prompt_template.format_messages(name="Bot", query=chat_input, context=context, history=st.session_state.memory.load_memory_variables({}))
    with st.chat_message("ai"):
        with st.spinner("Typing..."):
            content = ""
            with st.empty():
                for chunk in vertexAI.stream(prompt_value):
                    content += chunk.content
                    st.write(content)
            st.session_state.messages.append({"role": "ai", "content": content})

    st.session_state.memory.save_context({"input": chat_input}, {"output": content})

PGVector 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import ChatVertexAI
from langchain.prompts import ChatPromptTemplate
from langchain_google_vertexai import VertexAIEmbeddings
from langchain.memory import ConversationBufferWindowMemory
from langchain_community.vectorstores.pgvector import PGVector
import streamlit as st
import os

vertexAI = ChatVertexAI(model_name=os.getenv("VERTEX_AI_MODEL_NAME", "gemini-2.5-flash-preview-04-17"), streaming=True, convert_system_message_to_human=True)
prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant who helps in finding answers to questions using the provided context."),
        ("human", """
        The answer should be based on the text context given in "text_context" and the conversation history given in "conversation_history" along with its Caption: \n
        Base your response on the provided text context and the current conversation history to answer the query.
        Select the most relevant information from the context.
        Generate a draft response using the selected information. Remove duplicate content from the draft response.
        Generate your final response after adjusting it to increase accuracy and relevance.
        Now only show your final response!
        If you do not know the answer or context is not relevant, response with "I don't know".

        text_context:
        {context}

        conversation_history:
        {history}

        query:
        {query}
        """),
    ]
)

embedding_model = VertexAIEmbeddings("text-embedding-005")

CONNECTION_STRING = PGVector.connection_string_from_db_params(
    driver="psycopg2",
    host=os.environ.get("POSTGRES_HOST"),
    port=5432,
    database=os.environ.get("DATABASE_NAME"),
    user=os.environ.get("USERNAME"),
    password=os.environ.get("PASSWORD"),
)
COLLECTION_NAME = os.environ.get("COLLECTION_NAME"),

vector_search = PGVector(
    collection_name=COLLECTION_NAME,
    connection_string=CONNECTION_STRING,
    embedding_function=embedding_model,
)

def format_docs(docs):
    return "\n\n".join([d.page_content for d in docs])

st.title("🤖 Chatbot")
if "messages" not in st.session_state:
    st.session_state["messages"] = [{"role": "ai", "content": "How can I help you?"}]

if "memory" not in st.session_state:
    st.session_state["memory"] = ConversationBufferWindowMemory(
        memory_key="history",
        ai_prefix="Bot",
        human_prefix="User",
        k=3,
    )

for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.write(message["content"])

if chat_input := st.chat_input():
    with st.chat_message("human"):
        st.write(chat_input)
        st.session_state.messages.append({"role": "human", "content": chat_input})

    found_docs = vector_search.similarity_search(chat_input)
    context = format_docs(found_docs)

    prompt_value = prompt_template.format_messages(name="Bot", query=chat_input, context=context, history=st.session_state.memory.load_memory_variables({}))
    with st.chat_message("ai"):
        with st.spinner("Typing..."):
            content = ""
            with st.empty():
                for chunk in vertexAI.stream(prompt_value):
                    content += chunk.content
                    st.write(content)
            st.session_state.messages.append({"role": "ai", "content": content})

    st.session_state.memory.save_context({"input": chat_input}, {"output": content})

Weaviate 

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from langchain_google_vertexai import ChatVertexAI
from langchain.prompts import ChatPromptTemplate
from langchain_google_vertexai import VertexAIEmbeddings
from langchain.memory import ConversationBufferWindowMemory
import weaviate
from weaviate.connect import ConnectionParams
from langchain_weaviate.vectorstores import WeaviateVectorStore
import streamlit as st
import os

vertexAI = ChatVertexAI(model_name=os.getenv("VERTEX_AI_MODEL_NAME", "gemini-2.5-flash-preview-04-17"), streaming=True, convert_system_message_to_human=True)
prompt_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant who helps in finding answers to questions using the provided context."),
        ("human", """
        The answer should be based on the text context given in "text_context" and the conversation history given in "conversation_history" along with its Caption: \n
        Base your response on the provided text context and the current conversation history to answer the query.
        Select the most relevant information from the context.
        Generate a draft response using the selected information. Remove duplicate content from the draft response.
        Generate your final response after adjusting it to increase accuracy and relevance.
        Now only show your final response!
        If you do not know the answer or context is not relevant, response with "I don't know".

        text_context:
        {context}

        conversation_history:
        {history}

        query:
        {query}
        """),
    ]
)

embedding_model = VertexAIEmbeddings("text-embedding-005")

auth_config = weaviate.auth.AuthApiKey(api_key=os.getenv("APIKEY"))
client = weaviate.WeaviateClient(
    connection_params=ConnectionParams.from_params(
        http_host=os.getenv("WEAVIATE_ENDPOINT"),
        http_port="80",
        http_secure=False,
        grpc_host=os.getenv("WEAVIATE_GRPC_ENDPOINT"),
        grpc_port="50051",
        grpc_secure=False,
    ),
    auth_client_secret=auth_config
)
client.connect()

vector_search = WeaviateVectorStore.from_documents([],embedding_model,client=client, index_name="trainingdocs")

def format_docs(docs):
    return "\n\n".join([d.page_content for d in docs])

st.title("🤖 Chatbot")
if "messages" not in st.session_state:
    st.session_state["messages"] = [{"role": "ai", "content": "How can I help you?"}]

if "memory" not in st.session_state:
    st.session_state["memory"] = ConversationBufferWindowMemory(
        memory_key="history",
        ai_prefix="Bot",
        human_prefix="User",
        k=3,
    )

for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.write(message["content"])

if chat_input := st.chat_input():
    with st.chat_message("human"):
        st.write(chat_input)
        st.session_state.messages.append({"role": "human", "content": chat_input})

    found_docs = vector_search.similarity_search(chat_input)
    context = format_docs(found_docs)

    prompt_value = prompt_template.format_messages(name="Bot", query=chat_input, context=context, history=st.session_state.memory.load_memory_variables({}))
    with st.chat_message("ai"):
        with st.spinner("Typing..."):
            content = ""
            with st.empty():
                for chunk in vertexAI.stream(prompt_value):
                    content += chunk.content
                    st.write(content)
            st.session_state.messages.append({"role": "ai", "content": content})

    st.session_state.memory.save_context({"input": chat_input}, {"output": content})