Working with FLIP apps

A FLIP app is the user-supplied federated-learning code that runs across the Central Hub and each participating Trust. The guides in this section walk through how to adapt or build apps so they can be uploaded to FLIP and orchestrated by the FL framework (Flower or NVIDIA FLARE).

• Create a FLIP app from a Flower app
  • Starting point
  • ServerApp: reporting model lifecycle status
  • Strategy considerations
  • ClientApp: fetching the FLIP DataFrame
  • ClientApp: fetching images by accession
  • ClientApp: sending per-round metrics to the hub
  • Wiring it up via pyproject.toml
  • Submitting the app to FLIP
  • Local testing before upload
  • Common pitfalls
• Create a FLIP app from a FLARE app