Mini Map
Search
⌘ K
AI Features
Log In
Grokking the Generative AI System Design*
1.
Introduction to GenAI System Design
The New Hybrid Discipline
From Training to Reality
Step 1: The Heavy Lifting
Step 2: The System Design
Step 3: Real-World Delivery
Lab vs. Production
Why Design Matters?
Quick Recap
2.
Fundamental Concepts in GenAI
The Brain of AI
Evolution of Architectures
RNNs: The Memory Loop
Transformers: Attention is All You Need
How Attention Works
Tokens: Chopping Text
Embeddings as Meaning
Why We Need Many GPUs
Parallelism: Data vs Model
Making Models Faster
Speed Trick: Quantization
Speed Trick: Distillation
Choosing the Right Cache
Why Evaluation Is Hard
Quality Metrics Snapshot
Do Metrics Replace Humans?
GenAI Fundamentals Recap
3.
Back-of-the-Envelope Calculations
The Art of Estimation
FLOPS in One Step
Neural Network Parameters
What Is a Forward Pass?
What Is a Backward Pass?
FLOPS per Parameter
Training Time Formula
Quick Check: Training Factors
Estimating Model Storage
Precision vs. Size Tradeoff
Estimating Server Count
BOTEC Cheat Sheet
4.
Systematic Framework for Designing GenAI Systems
Taming GenAI Complexity
The SCALED Framework
System Requirements
Choose AI Model
Acquire Data
Leverage Model
Estimate Resources
Design System
Model Size Trade-offs
The Data Quality Rule
SCALED Recap
5.
System Design of a Text-to-Text Generation System
How Machines Understand Us
A Sample User Message
Spotting the User’s Goal
Picking Out Key Details
Reading the Emotion
Big vs. Small Models
Training Takes Time
Serving 100M Users
Three-Part AI Pipeline
Step 1: Prompt Processing
Step 2: Long-Term Memory
Step 3: Model and Safety
Vector Databases 101
Safety Check Quiz
Recap: Text-to-Text System
6.
System Design of a Text-to-Image Generation System
From Text to Pixels
The Winning Architecture: Diffusion
How Diffusion Works
Step 1: Forward Diffusion
Step 2: Backward Diffusion
Fueling the Model: Data
Training vs. Inference
Resource Reality Check
The Brain: Prompt Processing
Building the System Flow
Step 1: Understanding Intent
Step 2: Adding Context
Step 3: Generation and Safety
Recap: Text-to-Image System
7.
System Design of a Text-to-Speech Generation System
The TTS Revolution
Inside Fish Speech
Step 1: The Linguist
Step 2: The Composer
Step 3: The Performer
Fueling the Model
Designing the System
The Production Pipeline
Preprocessing
Phoneme Extraction
Voice Selection
Synthesis
The Cost of Scale
Pipeline Check
Recap: Text-to-Speech System
8.
System Design of a Text-to-Video Generation System
The 4th Dimension of AI
Inside the Mochi 1 Model
Dual Inputs
The Frozen Encoder
Multimodal Fusion
Evaluating Reality
Infrastructure Shock
The Deployment Pipeline
Phase 1: Prompt Processing
Phase 2: The Model Host
Phase 3: Rendering and QC
System Knowledge Check
Recap: Text-to-Video System
9.
System Design of an Image Captioning System
When Vision Meets Language
Anatomy of a VLM
The Eye: Image Encoder
The Bridge: Q-Former
The Voice: Language Decoder
The BLIP-2 Strategy
Grading the Caption
Inference Bottlenecks
Deployment Flow
Prep and Vision Stage
Generate With Context
Safety and Quality Checks
Recap: Image Captioning System
10.
Conclusion
The GenAI Architect
The Four Pillars of GenAI
Text Generation
Image Generation
Speech Generation
Video Generation
The Strategic Edge
Applying Your Toolkit
Claim your Certificate
Home
Courses
Grokking the Generative AI System Design*
The Bridge: Q-Former
This is a summary
We'll cover the following...
This component is not supported in the current context
...