Vertex AI

Workflow

• Vertex AI ML Workflow

Components

• Data sets
  • Managed by Vertex AI—can be linked to a model
• Feature store
  • Ingest, serve and share Machine Learning features
• Labelling tasks
  • Request human labelling of video, image or text data
• Vertex AI Workbench
  • Jupyter Notebook development environment
  • Can process data in Dataproc, train models etc.
  • Use the Vertex AI Python SDK
• Pipelines
  • Automate, monitor, govern ML systems
  • Orchestrate workflow
  • Store workflow artefacts using Vertex AI metadata—analyse lineage of workflow artefacts e.g. training data, hyperparameters, code etc.
• Training
  • AutoML
  • Custom training
• Experiments
  • Vertex Visia—optimization, tune hyperparameters
  • TensorBoard—compare studies
• Models
  • Manage models on Google Cloud
• Endpoints
  • Deploy models for serving predictions
  • Models trained in and out of Vertex AI
• Batch predictions
  • Group prediction requests
  • Asynchronous
• Metadata
  • Stores pipeline metadata

Generative AI

• Google Cloud for GenAI
• Gemini

Tools

• Connected to Cloud Logging and Cloud Monitoring
• Client libraries—native libraries (Python, Node.js, Java), and API libraries for other languages
• REST API
• Deep Learning VM Images—Compute Engine images with tools pre-installed e.g. TensorFlow, PyTorch
• Deep Learning Containers