Prompt Engineering concepts

• What is a Prompt? Prompts are inputs from the user that help guide LLMs on Amazon Bedrock to generate an appropriate response or output for a given task or instruction.
• What is Prompt Engineering? The practice of prompt engineering involves selecting the right words, phrases, and punctuation to maximize LLM usage in different applications. Prompt engineering is the art of communicating with an LLM. High-quality prompts condition the LLM to generate desired or better responses. The detailed guidance provided within this document is applicable across all LLMs within Amazon Bedrock.
• Few-shot prompting vs. zero-shot prompting
• Inference parameters – Values that can be adjusted during model inference to influence a response. Inference parameters can affect how varied responses are and can also limit the length of a response or the occurrence of specified sequences. For more information and definitions of specific inference parameters, see Influence response generation with inference parameters.

  

Data Preparation for Machine Learning

Amazon S3

• Get started managing partitions for Amazon S3 tables backed by the AWS Glue Data Catalog
• Building and optimizing a data lake on Amazon S3
• Set up AWS Glue Crawlers using S3 event notifications
• Understanding and managing Amazon S3 storage classes
• How Canva saves over $3 million annually in Amazon S3 costs?

  

Amazon Athena

• An interactive query service that makes it easy to analyze data in Amazon S3 using standard SQL. It is useful for querying datasets to prepare them for ML models
• Use AWS Glue to connect Athena to your data in Amazon S3
• Create a table from query results (CTAS)

Amazon Redshift

• Amazon Redshift Streaming Ingestion: Generate near real time insights through streaming data ingestion into your data warehouse and data visualizations
• Dynamic Data Masking(DDM) support in Amazon Redshift
• Build visualizations using Amazon Redshift query editor v2
• Amazon Redshift integration for Apache Spark

Amazon Comprehend

• Redact sensitive data from streaming data in near-real time using Amazon Comprehend and Amazon Kinesis Data Firehose
• Copying custom models between AWS accounts - With Amazon Comprehend, you can share your custom models with others, so they can import your models into their AWS accounts. When a user imports one of your custom models, they create a new custom model in their account. Their new model duplicates the one that you shared.
• You can import a custom model that's in another AWS account. When you import a model, you create a new custom model in your account. Your new custom model is a fully-trained duplicate of the model that you imported.

ML Model Development

• Amazon SageMaker - A comprehensive service that covers the entire machine learning workflow, including model building, training, tuning, and deployment. SageMaker supports built-in algorithms like Linear Learner and XGBoost, as well as custom algorithms.
• Amazon SageMaker: A deep dive
• Amazon SageMaker Inference explained: Which style is right for you?
• Built-in algorithms and pretrained models in Amazon SageMaker
• Hyperparameter Optimization - Understand strategies such as Hyperband, Bayesian optimization, and grid search to efficiently find the best model parameters
• SageMaker managed warm pools let you retain and reuse provisioned infrastructure after the completion of a training job to reduce latency for repetitive workloads, such as iterative experimentation or running many jobs consecutively. Subsequent training jobs that match specified parameters run on the retained warm pool infrastructure, which speeds up start times by reducing the time spent provisioning resources.

Deployment and Orchestration of ML Workflows

• Introduction to MLOps engineering on AWS Watch Download PDF
• End-to-end MLOps for architects Watch
• SageMaker Endpoints Learn about deploying models using real-time endpoints, serverless inference, and multi-model endpoints. This includes understanding how to set up CI/CD pipelines with AWS CodePipeline for automated deployment
• Use the SageMaker Model Registry for model versioning and governance, which helps manage different versions of models and track their lineage. Amazon SageMaker Collections is a new capability to organize models in the Model Registry.
• Centrally track and manage your model versions in Amazon SageMaker

Monitor and optimize infrastructure and costs

• Debug model output tensors from machine learning training jobs in real time and detect non-converging issues using Amazon SageMaker Debugger.
• Managed spot training can optimize the cost of training models up to 90% over on-demand instances. SageMaker manages the Spot interruptions on your behalf.
• Use checkpoints in Amazon SageMaker to save the state of machine learning (ML) models during training. Checkpoints are snapshots of the model and can be configured by the callback functions of ML frameworks. You can use the saved checkpoints to restart a training job from the last saved checkpoint.

Secure your AWS resources

• AWS Secrets Manager - Rotate secrets up to every four hours
• How to use AWS Secrets Manager to securely store and rotate SSH key pairs
• Guidance for Sensitive Information Scanning with Amazon Macie on AWS
• Protect your data using Amazon VPC and AWS PrivateLink
  - Use interface VPC endpoints (AWS PrivateLink) to create a private connection between your VPC and Amazon Bedrock
• Metrics for monitoring Amazon SageMaker with Amazon CloudWatch
• Log Amazon SageMaker API calls with AWS CloudTrail
• Well-Architected machine learning lens, Security best practices in Machine Learning Lens

Supplementary reading material

Wishing you the very best for your upcoming AWS Certified Machine Learning Engineer - Associate exam! If this blog post helped you prepare, please let me know in the comments.