GKE 및 Cloud Storage로 RAG 챗봇 빌드

Autopilot Standard

이 튜토리얼에서는 검색 증강 생성(RAG)을 기반으로 하는 대규모 언어 모델(LLM) 애플리케이션을 Cloud Storage 버킷에 업로드하는 PDF 파일과 통합하는 방법을 보여줍니다.

이 가이드에서는 데이터베이스를 업로드된 문서의 표현(임베딩)을 보유하는 스토리지 및 시맨틱 검색엔진으로 사용합니다. Langchain 프레임워크를 사용하여 임베딩과 상호작용하고 Vertex AI를 통해 제공되는 Gemini 모델을 사용합니다.

Langchain은 여러 머신러닝 태스크를 간소화하고 다양한 벡터 데이터베이스 및 AI 서비스와 통합할 수 있는 인터페이스가 있는 인기 있는 오픈소스 Python 프레임워크입니다.

이 튜토리얼은 GKE 및 Cloud Storage에 RAG LLM 애플리케이션을 배포하는 데 관심이 있는 클라우드 플랫폼 관리자 및 설계자, ML 엔지니어, MLOps(DevOps) 전문가를 대상으로 합니다.

목표

이 튜토리얼에서는 다음을 수행하는 방법을 알아봅니다.

벡터 데이터베이스에 문서 임베딩을 만들고 저장하는 애플리케이션을 빌드하고 배포합니다.
Cloud Storage 버킷에 새 문서 업로드를 트리거하도록 애플리케이션을 자동화합니다.
시맨틱 검색을 사용하여 문서 콘텐츠를 기반으로 질문에 답하는 챗봇 애플리케이션을 배포합니다.

배포 아키텍처

이 튜토리얼에서는 Cloud Storage 버킷, Eventarc 트리거와 다음 서비스를 만듭니다.

embed-docs: 사용자가 Cloud Storage 버킷에 새 문서를 업로드할 때마다 Eventarc가 이 서비스를 트리거합니다. 서비스는 업로드된 문서의 임베딩을 만들고 벡터 데이터베이스에 임베딩을 삽입하는 Kubernetes 작업을 시작합니다.
chatbot: 이 서비스는 시맨틱 검색 및 Gemini API를 사용하여 업로드된 문서에 관한 자연어 질문에 답변합니다.

다음 다이어그램은 문서를 업로드하고 벡터화하는 프로세스를 보여줍니다.

다이어그램에서 사용자는 Cloud Storage 버킷에 파일을 업로드합니다. Eventarc는 버킷의 객체 metadataUpdated 이벤트를 구독하고 Kubernetes 워크로드인 Eventarc의 이벤트 전달자를 사용하여 새 문서를 업로드할 때 embed-docs 서비스를 호출합니다. 그러면 서비스에서 업로드된 문서의 임베딩을 만듭니다. embed-docs 서비스는 Vertex AI 임베딩 모델을 사용하여 벡터 데이터베이스에 임베딩을 저장합니다.

다음 다이어그램은 chatbot 서비스를 사용하여 업로드된 문서 콘텐츠에 관해 질문하는 프로세스를 보여줍니다.

사용자는 자연어를 사용하여 질문할 수 있으며 챗봇은 업로드된 파일의 콘텐츠만을 기반으로 답변을 생성합니다. 챗봇은 시맨틱 검색을 사용하여 벡터 데이터베이스에서 컨텍스트를 가져온 다음 질문과 컨텍스트를 Gemini로 전송합니다.

비용

이 문서에서는 비용이 청구될 수 있는 다음과 같은 Google Cloud 구성요소를 사용합니다.

프로젝트 사용량을 기준으로 예상 비용을 산출하려면 가격 계산기를 사용하세요. Google Cloud를 처음 사용하는 사용자는 무료 체험판을 사용할 수 있습니다.

이 문서에 설명된 태스크를 완료했으면 만든 리소스를 삭제하여 청구가 계속되는 것을 방지할 수 있습니다. 자세한 내용은 삭제를 참조하세요.

시작하기 전에

이 튜토리얼에서는 Cloud Shell을 사용하여 명령어를 실행합니다. Cloud Shell은 Google Cloud에서 호스팅되는 리소스를 관리하는 데 사용되는 셸 환경입니다. Cloud Shell에는 Google Cloud CLI, kubectl, Terraform 명령줄 도구가 사전 설치되어 있습니다. Cloud Shell을 사용하지 않는 경우 Google Cloud CLI를 설치합니다.

Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

Install the Google Cloud CLI.

To initialize the gcloud CLI, run the following command:

gcloud init

Note: If you installed the gcloud CLI previously, make sure you have the latest version by running

gcloud components
      update

Create or select a Google Cloud project.

Create a Google Cloud project:
```
gcloud projects create PROJECT_ID
```
Replace PROJECT_ID with a name for the Google Cloud project you are creating.
Select the Google Cloud project that you created:
```
gcloud config set project PROJECT_ID
```
Replace PROJECT_ID with your Google Cloud project name.

Make sure that billing is enabled for your Google Cloud project.

Enable the Vertex AI, Cloud Build, Eventarc, Artifact Registry APIs:

gcloud services enable aiplatform.googleapis.com cloudbuild.googleapis.com eventarc.googleapis.com artifactregistry.googleapis.com

Install the Google Cloud CLI.

To initialize the gcloud CLI, run the following command:

gcloud init

Note: If you installed the gcloud CLI previously, make sure you have the latest version by running

gcloud components
      update

Create or select a Google Cloud project.

Create a Google Cloud project:
```
gcloud projects create PROJECT_ID
```
Replace PROJECT_ID with a name for the Google Cloud project you are creating.
Select the Google Cloud project that you created:
```
gcloud config set project PROJECT_ID
```
Replace PROJECT_ID with your Google Cloud project name.

Make sure that billing is enabled for your Google Cloud project.

Enable the Vertex AI, Cloud Build, Eventarc, Artifact Registry APIs:

gcloud services enable aiplatform.googleapis.com cloudbuild.googleapis.com eventarc.googleapis.com artifactregistry.googleapis.com

Grant roles to your user account. Run the following command once for each of the following IAM roles: eventarc.admin
```
gcloud projects add-iam-policy-binding PROJECT_ID --member="user:USER_IDENTIFIER" --role=ROLE
```
- Replace PROJECT_ID with your project ID.
- Replace USER_IDENTIFIER with the identifier for your user account. For example, user:myemail@example.com.
- Replace ROLE with each individual role.

클러스터 만들기

Qdrant, Elasticsearch 또는 Postgres 클러스터를 만듭니다.

Qdrant

GKE에 Qdrant 벡터 데이터베이스 배포의 안내에 따라 Autopilot 모드 또는 Standard 모드 GKE 클러스터에서 실행되는 Qdrant 클러스터를 만듭니다.

Elasticsearch

GKE에 Elasticsearch 벡터 데이터베이스 배포의 안내에 따라 Autopilot 모드 또는 Standard 모드 GKE 클러스터에서 실행되는 Elasticsearch 클러스터를 만듭니다.

PGVector

GKE에 PostgreSQL 벡터 데이터베이스 배포의 안내에 따라 Autopilot 모드 또는 Standard 모드 GKE 클러스터에서 PGVector가 실행되는 Postgres 클러스터를 만듭니다.

Weaviate

안내에 따라 GKE에 Weaviate 벡터 데이터베이스를 배포하여 Autopilot 또는 Standard 모드 GKE 클러스터에서 실행되는 Weaviate 클러스터를 만듭니다.

환경 설정하기

Cloud Shell로 환경을 설정합니다.

프로젝트의 환경 변수를 설정합니다.

Qdrant

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

PROJECT_ID를 Google Cloud 프로젝트 ID로 바꿉니다.

Elasticsearch

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

PROJECT_ID를 Google Cloud 프로젝트 ID로 바꿉니다.

PGVector

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

PROJECT_ID를 Google Cloud 프로젝트 ID로 바꿉니다.

Weaviate

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

PROJECT_ID를 Google Cloud 프로젝트 ID로 바꿉니다.

GKE 클러스터 실행 여부를 확인합니다.

gcloud container clusters list --project=${PROJECT_ID} --region=${REGION}

출력은 다음과 비슷합니다.

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

GitHub에서 샘플 코드 저장소를 클론합니다.

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

databases 디렉터리로 이동합니다.
```
cd kubernetes-engine-samples/databases
```

인프라 준비

Artifact Registry 저장소를 만들고 Docker 이미지를 빌드한 후 Docker 이미지를 Artifact Registry에 푸시합니다.

Artifact Registry 저장소를 만듭니다.

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

Compute Engine 서비스 계정에 storage.objectAdmin 및 artifactregistry.admin 권한을 설정하여 Cloud Build를 사용하여 embed-docs 및 chatbot 서비스의 Docker 이미지를 빌드하고 푸시합니다.

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"

PROJECT_NUMBER를 Google Cloud 프로젝트 번호로 바꿉니다.

embed-docs 및 chatbot 서비스의 Docker 이미지를 빌드합니다. embed-docs 이미지에는 Eventarc 전달자 요청을 수신하는 애플리케이션과 임베딩 작업 모두에 관한 Python 코드가 포함되어 있습니다.

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

이미지를 확인합니다.

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

출력은 다음과 비슷합니다.

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

Kubernetes 작업을 실행할 권한이 있는 Kubernetes 서비스 계정을 배포합니다.

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 -

Terraform을 사용하여 GKE 클러스터를 만들고 create_service_account를 true로 설정하면 클러스터와 노드에서 별도의 서비스 계정이 생성되고 사용됩니다. 이 Compute Engine 서비스 계정에 artifactregistry.serviceAgent 역할을 부여하여 노드가 embed-docs 및 chatbot용으로 생성된 Artifact Registry에서 이미지를 가져올 수 있도록 합니다.
```
export CLUSTER_SERVICE_ACCOUNT=$(gcloud container clusters describe ${KUBERNETES_CLUSTER_PREFIX}-cluster \
--region=${REGION} \
--format="value(nodeConfig.serviceAccount)")

gcloud projects add-iam-policy-binding ${PROJECT_ID}  \
--member="serviceAccount:${CLUSTER_SERVICE_ACCOUNT}" \
--role="roles/artifactregistry.serviceAgent"
```
서비스 계정에 액세스 권한을 부여하지 않으면 embed-docs 및 chatbot 서비스를 배포할 때 Artifact Registry에서 이미지를 가져오려고 할 때 노드에 권한 문제가 발생할 수 있습니다.

embed-docs 및 chatbot 서비스용 Kubernetes 배포를 배포합니다.

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 -

GKE에 Eventarc 트리거를 사용 설정합니다.
```
gcloud eventarc gke-destinations init
```
메시지가 표시되면 y를 입력합니다.
Cloud Storage 버킷을 배포하고 Terraform을 사용하여 Eventarc 트리거를 만듭니다.
```
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}
```
메시지가 표시되면 yes를 입력합니다. 명령어가 완료되는 데 몇 분 정도 걸릴 수 있습니다.

Terraform에서 다음 리소스를 만듭니다.
- 문서를 업로드할 Cloud Storage 버킷
- Eventarc 트리거
- Eventarc를 사용할 권한이 있는 service_account_eventarc_name이라는 Google Cloud 서비스 계정
- 버킷을 읽고 Vertex AI 모델에 액세스할 수 있는 권한이 있는 service_account_bucket_name이라는 Google Cloud 서비스 계정
출력은 다음과 비슷합니다.
```
... # Several lines of output omitted

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

... # Several lines of output omitted
```

문서 로드 및 챗봇 쿼리 실행

데모 문서를 업로드하고 쿼리를 실행하여 챗봇을 사용하여 데모 문서를 검색합니다.

버킷에 예시 carbon-free-energy.pdf 문서를 업로드합니다.

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

문서 삽입기 작업이 성공적으로 완료되었는지 확인합니다.

kubectl get job -n ${DB_NAMESPACE}

출력은 다음과 비슷합니다.

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

부하 분산기의 외부 IP 주소를 가져옵니다.

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

웹브라우저에서 외부 IP 주소를 엽니다.
```
http://EXTERNAL_IP
```
챗봇은 다음과 유사한 메시지로 응답합니다.
```
How can I help you?
```

업로드된 문서의 콘텐츠에 관해 질문합니다. 챗봇이 아무것도 찾을 수 없는 경우 I don't know로 응답합니다. 예를 들어 다음과 같이 질문할 수 있습니다.

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

챗봇의 출력 예시는 다음과 같습니다.

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.

업로드된 문서의 맥락과 관련 없는 내용을 챗봇에 질문합니다. 예를 들어 다음과 같이 질문할 수 있습니다.
```
You: What are Google plans to colonize Mars?
```
챗봇의 출력 예시는 다음과 같습니다.
```
Bot: I don't know. The provided context does not mention anything about Google's plans to colonize Mars.
```

애플리케이션 코드 정보

이 섹션에서는 애플리케이션 코드가 작동하는 방식을 설명합니다. Docker 이미지 내부에는 세 가지 스크립트가 있습니다.

endpoint.py: 각 문서 업로드 시 Eventarc 이벤트를 수신하고 이를 처리하기 위한 Kubernetes 작업을 시작합니다.
embedding-job.py: 버킷에서 문서를 다운로드하고, 임베딩을 생성하고, 벡터 데이터베이스에 임베딩을 삽입합니다.
chat.py: 저장된 문서의 콘텐츠에 대해 쿼리를 실행합니다.

다음 다이어그램은 문서 데이터를 사용하여 답변을 생성하는 과정을 보여줍니다.

다이어그램에서 애플리케이션은 PDF 파일을 로드하고 파일을 청크로 분할한 다음 벡터로 변환한 후 벡터를 벡터 데이터베이스로 전송합니다. 나중에 사용자가 챗봇에 질문합니다. RAG 체인은 시맨틱 검색을 사용하여 벡터 데이터베이스를 검색한 후 질문과 함께 컨텍스트를 LLM에 반환합니다. LLM은 질문에 답변하고 질문을 채팅 기록에 저장합니다.

`endpoint.py` 정보

이 파일은 Eventarc의 메시지를 처리하고, 문서를 임베딩하기 위한 Kubernetes 작업을 만들고, 포트 5001에서 어디서나 요청을 수락합니다.

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)

`embedding-job.py` 정보

이 파일은 문서를 처리하고 벡터 데이터베이스로 전송합니다.

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="gemini-pro", 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("textembedding-gecko@001")

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("textembedding-gecko@001")

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("textembedding-gecko@001")

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("textembedding-gecko@001")

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)}")

`chat.py` 정보

이 파일은 제공된 컨텍스트와 이전 답변만 사용하여 질문에 답변하도록 모델을 구성합니다. 컨텍스트 또는 대화 기록이 데이터와 일치하지 않으면 모델은 I don't know를 반환합니다.

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="gemini-pro", 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("textembedding-gecko@001")

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="gemini-pro", 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("textembedding-gecko@001")

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="gemini-pro", 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("textembedding-gecko@001")

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})

삭제

이 튜토리얼에서 사용된 리소스 비용이 Google Cloud 계정에 청구되지 않도록 하려면 리소스가 포함된 프로젝트를 삭제하거나 프로젝트를 유지하고 개별 리소스를 삭제하세요.

프로젝트 삭제

청구되지 않도록 하는 가장 쉬운 방법은 튜토리얼에서 만든 프로젝트를 삭제하는 것입니다.

주의: 프로젝트를 삭제하면 다음과 같은 효과가 발생합니다.

프로젝트의 모든 항목이 삭제됩니다. 이 문서의 태스크에 기존 프로젝트를 사용한 경우 프로젝트를 삭제하면 프로젝트에서 수행한 다른 작업도 삭제됩니다.
커스텀 프로젝트 ID가 손실됩니다. 이 프로젝트를 만들 때 앞으로 사용할 커스텀 프로젝트 ID를 만들었을 수 있습니다. appspot.com URL과 같이 프로젝트 ID를 사용하는 URL을 보존하려면 전체 프로젝트를 삭제하는 대신 프로젝트 내에서 선택한 리소스만 삭제합니다.

여러 아키텍처, 튜토리얼, 빠른 시작을 살펴보려는 경우 프로젝트를 재사용하면 프로젝트 할당량 한도 초과를 방지할 수 있습니다.

Delete a Google Cloud project:

gcloud projects delete PROJECT_ID

프로젝트를 삭제하면 정리가 완료됩니다. 프로젝트를 삭제하지 않은 경우 개별 리소스 삭제를 진행합니다.

개별 리소스 삭제

Artifact Registry 저장소를 삭제합니다.

gcloud artifacts repositories delete ${KUBERNETES_CLUSTER_PREFIX}-images \
    --location=${REGION} \
    --async

메시지가 표시되면 y를 입력합니다.

Cloud Storage 버킷과 Eventarc 트리거를 삭제합니다.

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

메시지가 표시되면 yes를 입력합니다.

Eventarc를 사용하려면 생성 시와 삭제 시 모두 유효한 엔드포인트 타겟이 있어야 합니다.

다음 단계

GKE의 데이터베이스 배포 권장사항 알아보기
GKE를 사용하여 데이터 집약적인 워크로드를 실행하기 위한 솔루션 살펴보기

GKE 및 Cloud Storage로 RAG 챗봇 빌드

목표

배포 아키텍처

비용

시작하기 전에

클러스터 만들기

Qdrant

Elasticsearch

PGVector

Weaviate

환경 설정하기

Qdrant

Elasticsearch

PGVector

Weaviate

인프라 준비

Qdrant

Elasticsearch

PGVector

Weaviate

Qdrant

Elasticsearch

PGVector

Weaviate

Qdrant

Elasticsearch

PGVector

Weaviate

문서 로드 및 챗봇 쿼리 실행

애플리케이션 코드 정보

endpoint.py 정보

Qdrant

Elasticsearch

PGVector

Weaviate

embedding-job.py 정보

Qdrant

Elasticsearch

PGVector

Weaviate

chat.py 정보

Qdrant

Elasticsearch

PGVector

Weaviate

삭제

프로젝트 삭제

개별 리소스 삭제

다음 단계

`endpoint.py` 정보

`embedding-job.py` 정보

`chat.py` 정보