AI Workload Documentation

AI workloads use non-deterministic behavior, making them more adaptable to complex situations compared to traditional workloads. They combine code and data into models, creating unique experiences. This guidance focuses on the architectural challenges of designing AI workloads, including non-deterministic functionality, data and application design, and operations.

Get started

Overview

• What is an AI workload?

Concept

• Design methodology
• Design principles
• Workload personas

Technical design areas

Concept

• Application design
• Application platform
• Training data design
• Grounding data design
• Data platform

Operational design areas

Concept

• Workload operations
• MLOps and GenAIOps
• Testing and evaluation
• Responsible AI

Reference examples

Architecture

• Baseline OpenAI end-to-end chat reference architecture

Related services

Architecture

• Azure Machine Learning service guide
• Azure OpenAI service guide

Assessment

How-To Guide

• AI assessment tool