API service for generating text and image embeddings using CLIP B/32 and L/14 models deployed on AWS SageMaker.

• Two API Endpoints: Separate API Gateway endpoints are deployed—one for the CLIP ViT-B/32 model and another for the ViT-L/14 model—allowing flexible embedding generation.

• Image & Text Embedding: Supports both image and text inputs to produce corresponding embeddings.

• AWS SageMaker Deployment: Models are deployed on AWS SageMaker serverless endpoints for scalable inference.

• Lambda Integration: AWS Lambda functions handle API requests and interface with the appropriate SageMaker endpoints.

• Monitoring: CloudWatch Alarms and SNS topics are set up for monitoring model health and performance.

• CLIP-ViT-B-32/
  Contains deployment scripts and configurations for the ViT-B/32 model.

• CLIP-ViT-L-14/
  Contains deployment scripts and configurations for the ViT-L/14 model.

• lambda_code_b32model.py
  AWS Lambda function code for handling requests to the B/32 model sagemaker endpoint.

• lambda_code_l14model.py
  AWS Lambda function code for handling requests to the L/14 model sagemaker endpoint.

• clip-ViT-B-32_deploy_job.ipynb
  Jupyter notebook for deploying the ViT-B/32 model to SageMaker.

• clip-ViT-L-14_deploy_job.ipynb
  Jupyter notebook for deploying the ViT-L/14 model to SageMaker.

Use the provided Jupyter notebooks to deploy the CLIP models to AWS SageMaker endpoints:

• clip-ViT-B-32_deploy_job.ipynb
• clip-ViT-L-14_deploy_job.ipynb

Deploy the Lambda functions to handle API requests:

• lambda_code_b32model.py — for the ViT-B/32 model
• lambda_code_l14model.py — for the ViT-L/14 model

Configure Amazon API Gateway to route requests to the appropriate Lambda functions.
Each model has its own dedicated API Gateway endpoint.

Set up CloudWatch Alarms and SNS topics to monitor the health and performance of the models.
Ensure notification subscriptions are confirmed to receive alerts.

Send POST requests to the API Gateway endpoints with either text or image data.

{
  "text": "Your input text here"
}

{
  "image": "Base64-encoded image string"
}

The API returns a JSON array containing the normalized embedding vector:

[
  0.057369355573432444,
  0.03234559691991972,
  ...
]