DS1

Appearance

SageMaker Notebook: DS1 Model Deployment Guide ​

Overview ​

This Jupyter notebook provides a comprehensive guide for deploying DS1 model packages from the AWS Marketplace using Amazon SageMaker. It demonstrates two primary deployment methods for machine learning inference workloads.

Purpose ​

The notebook walks users through the complete process of subscribing to, configuring, and deploying DS1 embedding models on AWS infrastructure, with support for both on-demand and batch processing scenarios.

Major Topics Covered ​

1. Deployment Methods ​

Real-Time Inference Endpoints

  • Persistent, fully managed APIs for on-demand inference
  • Handles individual requests with low latency
  • Suitable for production applications requiring immediate responses

Batch Transform Jobs

  • Finite execution runs for processing large datasets
  • Outputs predictions to files in bulk
  • Cost-effective for processing data at scale

2. Prerequisites and Setup ​

  • AWS Marketplace subscription process for DS1 model packages
  • AWS CLI SSO configuration for secure authentication
  • SageMaker quota verification for target regions
  • Instance type selection and compatibility checking

3. Model Configuration ​

  • Model ID specification (DS1-EN-V1)
  • Instance type selection for different deployment methods
  • Compute-optimised instances (ml.c5 family) recommended as DS1 doesn't require GPU
  • Amazon Resource Name (ARN) management across multiple AWS regions

4. AWS Integration ​

  • Boto3 session initialisation with SSO authentication
  • SageMaker runtime client setup
  • IAM role and permission management
  • Multi-region support configuration

5. Real-Time Endpoint Deployment ​

  • Model deployment to SageMaker endpoints
  • Endpoint configuration and initialisation
  • Testing with sample documents
  • Endpoint invocation and response handling
  • Endpoint lifecycle management (creation and deletion)

6. Batch Transform Jobs ​

  • S3 bucket setup for input/output data
  • JSONL input file format preparation
  • Each record contains document text and unique batch request ID
  • Transformer initialisation with instance configuration
  • Batch job execution and monitoring
  • Output matching with source documents using request IDs

7. Data Handling ​

  • Sample document preparation for testing
  • Input data formatting for both deployment methods
  • Response parsing and embedding extraction
  • Statistical analysis of embedding outputs (mean, standard deviation)

Technical Requirements ​

  • Python environment with boto3 and sagemaker libraries
  • AWS credentials configured via SSO or environment variables
  • Sufficient SageMaker quota in target AWS region
  • S3 bucket access for batch transform operations

Key Features ​

  • Support for multiple AWS regions across US, EU, APAC, and SA
  • Detailed error handling and authentication validation
  • Comprehensive output formatting and analysis
  • Cost-optimised instance recommendations
  • Production-ready deployment patterns

Use Cases ​

This notebook is ideal for teams looking to:

  • Deploy text embedding models for semantic search applications
  • Process large document collections in batch mode
  • Build real-time embedding services for production applications
  • Integrate ML models from AWS Marketplace into existing workflows