Skip to main content

Serving Models

Model-as-a-Service

  • Access model over internet
  • Surface model through web portal
  • Users post queries/data and receive predictions

On-Device Serving

  • Edge computing
  • Helpful for unreliable internet or secure environments

Pros of On-Device Serving:

  • Lower latency
  • Enhanced security
  • Works in remote/disconnected areas

Cons of On-Device Serving:

  • Resource constraints
  • Challenges with model updates
  • Monitoring difficulties

Implementation Strategies:

  • Model pruning
  • Transfer learning
  • Use of dedicated frameworks for edge deployment