Google Cloud Generative AI - Self-Directed Study Track Flashcards¶

81 cards from this guide's fact sheet and notes. Click a question to reveal the answer.

To study with spaced repetition, download flashcards.csv and import it into Anki or Quizlet (comma separated, first row is the header). Back to the study guide.

Generative ai fundamentals¶

Discriminative AI (generative ai fundamentals)

Classifies, predicts (Y given X)

Generative AI (generative ai fundamentals)

Creates new data (generates X)

Gecko (generative ai fundamentals)

Lightweight, mobile/IoT deployment

Otter (generative ai fundamentals)

Standard tasks, balanced performance

Bison (generative ai fundamentals)

Enhanced capabilities, most common

Unicorn (generative ai fundamentals)

Most capable, complex reasoning

Gemini Ultra (generative ai fundamentals)

Most capable, complex tasks

Gemini Pro (generative ai fundamentals)

Best performance/cost balance

Gemini Nano (generative ai fundamentals)

On-device, efficient

code-bison (generative ai fundamentals)

Code generation from natural language

code-gecko (generative ai fundamentals)

Code completion and suggestions

codechat-bison (generative ai fundamentals)

Conversational code assistance

Top-k (generative ai fundamentals)

Consider top k tokens

Stop sequences (generative ai fundamentals)

Custom stopping conditions

BLOCK_ONLY_HIGH (generative ai fundamentals)

Block high-probability harmful content

BLOCK_MEDIUM_AND_ABOVE (generative ai fundamentals)

Block medium+ harmful content (default)

BLOCK_LOW_AND_ABOVE (generative ai fundamentals)

Block low+ harmful content (most restrictive)

Vertex ai¶

Unified platform (vertex ai)

One place for all ML workflows

MLOps integration (vertex ai)

Built-in experiment tracking and model management

Scalable infrastructure (vertex ai)

Automatic scaling and optimization

Pre-built solutions (vertex ai)

AutoML and pre-trained models

Custom development (vertex ai)

Full flexibility for custom models

Managed notebooks (vertex ai)

Fully managed Jupyter instances

User-managed notebooks (vertex ai)

More control over environment

Built-in libraries (vertex ai)

Pre-installed ML frameworks

Data science tools (vertex ai)

Integrated development experience

Version control (vertex ai)

Git integration

Custom training (vertex ai)

Your own training code

AutoML training (vertex ai)

Automated model development

Hyperparameter tuning (vertex ai)

Automated hyperparameter optimization

Distributed training (vertex ai)

Multi-GPU and multi-node training

Online prediction (vertex ai)

Real-time inference

Batch prediction (vertex ai)

Offline batch processing

Private endpoints (vertex ai)

VPC-native serving

Multi-model endpoints (vertex ai)

Host multiple models

Components (vertex ai)

Reusable pipeline steps

Experiments (vertex ai)

Track and compare pipeline runs

Artifacts (vertex ai)

Track inputs, outputs, and metadata

Lineage (vertex ai)

Data and model lineage tracking

Model versioning (vertex ai)

Track model versions

Model lineage (vertex ai)

Understand model ancestry

Model evaluation (vertex ai)

Compare model performance

Model aliases (vertex ai)

Stable references to models

Text generation (vertex ai)

Create and edit text content

Chat (vertex ai)

Conversational AI applications

Code generation (vertex ai)

Programming assistance

Multimodal (vertex ai)

Text and image understanding

Prompt design (vertex ai)

Test and iterate on prompts

Model tuning (vertex ai)

Fine-tune foundation models

PaLM 2 (vertex ai)

text-bison, chat-bison, code-bison

Gemini (vertex ai)

gemini-pro, gemini-pro-vision, gemini-ultra

Imagen (vertex ai)

imagen-v1, imagen-v2

Chirp (vertex ai)

Speech recognition

BERT (vertex ai)

Text embeddings

T5 (vertex ai)

Text-to-text generation

Meta (vertex ai)

Llama 2, Llama 3 (7B, 13B, 70B parameters)

Anthropic (vertex ai)

Claude 2, Claude 3

Cohere (vertex ai)

Command, Embed, Rerank

AI21 (vertex ai)

Jurassic-2 models

Stability AI (vertex ai)

Stable Diffusion XL

CodeLlama (vertex ai)

Code generation

Classification (vertex ai)

Predict categories

Regression (vertex ai)

Predict numerical values

Feature engineering (vertex ai)

Automatic feature selection

Model selection (vertex ai)

Algorithm selection and tuning

Image classification (vertex ai)

Categorize images

Object detection (vertex ai)

Detect and locate objects

Image segmentation (vertex ai)

Pixel-level image understanding

Text classification (vertex ai)

Categorize text documents

Entity extraction (vertex ai)

Identify and extract entities

Sentiment analysis (vertex ai)

Analyze text sentiment

Question (vertex ai)

"You need to analyze images and generate descriptions. Which model?"

Answer (vertex ai)

Gemini Pro Vision - multimodal capabilities

Why Not (vertex ai)

text-bison (text only), code-bison (code focused)

Additional (vertex ai)

Lower temperature (0.1-0.3), provide context, use RAG

Prompt Engineering for (vertex ai)

Few examples available

Foundation Models (vertex ai)

PaLM 2, Gemini, embeddings

Tuning (vertex ai)

Supervised, adapter-based, RLHF

Grounding (vertex ai)

Search, RAG, custom sources

Enterprise (vertex ai)

Security, compliance, private deployment

Cost Effective (vertex ai)

Multiple pricing tiers, optimization options