Customizing and Evaluating LLMs Using Amazon SageMaker JumpStart

In this course, you learn about customizing and evaluating large language models (LLMs) using Amazon SageMaker JumpStart. Amazon SageMaker JumpStart is a machine learning (ML) hub with foundation models, built-in algorithms, and prebuilt ML solutions that you can deploy with a few clicks. You will learn the alternatives to fine-tuning including the foundations of prompt engineering and retrieval augmented generation (RAG). You will also learn to fine-tune, deploy and evaluate fine-tuned models available on SageMaker JumpStart.

Using your own AWS account and the notebooks provided, you can practice building RAG applications using the Amazon SageMaker-LangChain integration. You can also fine-tune a Llama3 model and evaluate it using evaluation metrics. You can practice one of the aspects of responsible AI with the help of a notebook that addresses prompt stereotyping. Alternatively, you can watch a video demonstration of running the notebooks.

• Course level: Advanced
• Duration: 4 hours

Activities

This course includes presentations, demonstrations, and assessments.

Course objectives

In this course, you will do the following:

• Describe the different techniques to customize LLMs.
• Describe when to use prompt engineering and retrieval augmented generation as customization options.
• Demonstrate the use of Amazon SageMaker-LangChain integration to build a RAG application using a Falcon model.
• Describe the use of domain adaptation and instruction fine-tuning.
• Demonstrate how to fine-tune and deploy a model from the SageMaker JumpStart ML hub.
• Demonstrate the use of the SageMaker Python SDK to fine-tune LLMs using Parameter Efficient Fine-Tuning (PEFT).
• Evaluate foundation models by using the SageMaker JumpStart console and fmeval library.

Intended Audience:

This course is intended for the following job roles:

• Data scientists
• Machine learning engineers

Prerequisites

We recommend that attendees of this course have the following:

• More than 1 year of experience with natural language processing (NLP)
• More than 1 year of experience with training and tuning language models
• Intermediate-level proficiency in Python language programming
• AWS Technical Essentials.
• Amazon SageMaker JumpStart Foundations.

Course outline

• Module 1: Introduction to Customizing LLMs
  • Customizing LLMs
  • Choosing customization methods
• Module 2: Prompt Engineering and RAG for Customizing LLMs
  • Using prompt engineering
  • Using Retrieval Augmented Generation (RAG)
  • Using advanced RAG patterns.
• Demo 1: Create a RAG application using Amazon SageMaker-LangChain integration and a Falcon 7B model from SageMaker JumpStart
• Module 3: Fine-tuning and Deploying Foundation Models
  • Customize foundation models using fine tuning
  • How to use SageMaker JumpStart console to fine-tune and deploy an LLM
• Demo 2: Fine-tune a Llama 3 model available on SageMaker JumpStart using Amazon SageMaker Python SDK
• Module 4: Evaluating Foundation Models
  • Discuss model evaluation metrics
  • Evaluate foundations models using Amazon SageMaker JumpStart console
• Demo 3: Evaluate prompt stereotyping of a Falcon-7B model using the fmeval library
• Module 5: Resources
  • Learn More
  • Contact Us