Skip to content

Latest commit

 

History

History
111 lines (85 loc) · 5.03 KB

README.md

File metadata and controls

111 lines (85 loc) · 5.03 KB

Serverless dbless RAG

Serverless RAG is an implementation of Retrieval Augumented Generation (RAG) on AWS. This solution is composed of AWS CDK resources and React project. Architecture is mainly composed of serverless, event-driven AWS services to deliver RAG experience for uses with low costs.

日本語の説明はこちら 日本語

Demo

serverless-dbless-rag-demo.mp4

Architecture

Architecture

Deploying the Solution

Prerequisites

  • AWS CDK
  • Docker
  • Node.JS
  • Enable the following model access on your AWS Account:
    • Amazon Titan Embed Text V1
    • Claude 3 Haiku

QuickStart

You can start the DBless Serverless RAG by executing the following commands:

git clone [email protected]:aws-samples/sample-serverless-dbless-rag-on-aws.git
cd sample-serverless-dbless-rag-on-aws
cd cdk
npm install
cdk deploy --context createFrontend=true --context generateInitialUser=true --context enableSnapStart=true

You can access the deployed CloudFront URL and log in using the username and initial password output during deployment. When you upload documents (PDF) through the GUI, Lambda will automatically process the embedding jobs sequentially via SQS. After the embedding is complete, you can perform Q&A on the content of the embedded documents through the GUI.

Enabling/Disabling Lambda SnapStart

You can enable Lambda SnapStart to improve UX by reducing the impact of cold starts with the following option:

cdk deploy --context enableSnapStart=true

Deploying with/without Frontend

You can deploy including the frontend by setting the createFrontend context to true:

cdk deploy --context createFrontend=true

When the createFrontend context is true, you can set up the initial user for the frontend by setting the generateInitialUser context to true:

cdk deploy --context createFrontend=true --context generateInitialUser=true

You can access the GUI for document embedding and search by accessing the deployed CloudFront URL.

Deploying without Frontend

cdk deploy

You can manually invoke the Lambda function as follows:

aws lambda invoke --function-name <Function Name> \
--cli-binary-format raw-in-base64-out \
--payload '{ "question": "What is EC2 instance?" }' output.txt

Estimated Costs

Estimated costs when this solution is deployed in the ap-northeast-1 region are as follows.

Note: Minor cost items and GUI costs are excluded from the table below.

Cost for 1000 Question Answering sessions

Service Item Quantity Cost (USD)
Lambda(x86, 2GB) GB-Duration (sec) 2 * 1000 0.03333
Lambda Number of requests 1000 0.0002
Bedrock Claude 3 Haiku, input tokens 1000 * 1000 0.25
Bedrock Claude 3 Haiku, output tokens 300 * 1000 0.375
Total estimated cost - - 0.65373

Cost for Embedding 1 MB of documents

Service Item Quantity Cost (USD)
Lambda(x86, 4GB) GB-Duration (sec) 4 * 20 0.0133
Bedrock Amazon Titan Text Embeddings, input tokens 30000 0.006
Total estimated cost - - 0.0143

Monthly Storage Costs

Service Item Quantity Cost (USD)
ECR Image size (GB) 1 0.1
S3 Assets size(GB) 1 0.025

Notes:

  • The actual costs may vary depending on the documents being embedded and the search content. This information is provided for reference only.
  • Some AWS services offer a free tier. This calculation does not take into account the free tier, so you may be able to use the services at a lower cost in practice.
  • This calculation is based on the pricing in the Tokyo region. While we strive to keep this information up-to-date, if there is any discrepancy between the prices listed here and those on the official AWS website, the prices on the official AWS website take precedence.

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.