Text Generation Gen AI with Vertex

(C) Copyright Elephant Scale February 27, 2024

Course Description

  • Large Language Models (LLM) are taking the world by storm. It can be ChatGPT, Claude, or Bard, but AI is doubling productivity in many occupations.
  • This course covers AI and Generative AI landscape, using Google Cloud’s Vertex AI.

Audience

  • Developers, data scientists, team leads, project managers

Skill Level

  • Introductory for all

Duration

  • One or two days

Prerequisites

  • General familiarity with machine learning

Format

  • Lectures and hands-on labs. (50% – 50%)

Lab environment

  • Zero Install: There is no need to install software on students’ machines!
  • A lab environment in the cloud will be provided for students.

Students will need the following

  • A reasonably modern laptop with unrestricted connection to the Internet. Laptops with overly restrictive VPNs or firewalls may not work properly.
    • A checklist to verify connectivity will be provided
  • Chrome browser

Generative AI on Vertex AI

  • Vertex AI on Google Cloud
  • Generative AI options on Google Cloud
  • Introduction to the Course Use Case (Text Generation)

Vertex AI Studio

  • Introduction to Vertex AI Studio
  • Available models and use cases
  • Designing and testing prompts in the Google Cloud console
  • Data governance in Vertex AI Studio
  • Lab: Getting Started with Vertex AI Studio’s User Interface

Prompt Design

  • Why is prompt design so important?
  • Zero-shot vs. few-shot prompting
  • Providing additional context and instruction-tuning
  • Best practices
  • Lab: Question Answering with Generative Models on Vertex AI

Implementing the PaLM API

  • Lab: Getting Started with the Vertex AI PaLM API and Python SDK

Introduction to the PaLM API

  • Utilizing generative models using the Python SDK
  • Understanding model parameters for text generation
  • Lab: Use the PaLM API to Integrate GenAI into Applications

Fine-Tuning Models

  • Scenarios to use model tuning
  • Workflow for model tuning
  • Preparing your model-tuning dataset
  • Create a model-tuning job
  • Loading a tuned model
  • Demo: Fine-Tuning Models for Your Specific Use Case