Release Notes

v0.2.5

What’s new?

Other

  • Documentation updated

  • Introduce Discord community server

v0.2.4

What’s new?

  • Introduced a new events view.

  • Introduced a new view for viewing network layout, (reducer, combiner and clients hierarchy)

  • Introduced a new setup guide-phase to ensure prereqs like package and model are set before starting execution.

  • Introduced a better form for parameter selection on run configuration.

  • Introduced async dispatching of run configurations.

  • Introduced async update refresh of several important fields for user convenincence like status, events, network hierarchy etc.

  • Introduced a new download-client-config function to allow for faster and more convenient client configuration. (Just download config and point your local client and whoallah! You are online in this federation.)

Other

  • Fixed logic bugs related to framework persistance.

  • Fixed a logic bug causing clients to get assigned prior to compute package assignment (and hence will not account for assignment policy).

  • Fixed a logic bug if reducer is resumed from previous state (to ensure) that the right compute package is selected.

  • Update dependency versions.

v0.2.3

What’s new?

  • Support for latest Minio

  • Improvements i UI - now not possible to submit jobs is in monitoring state.

  • Improvement of Docker image hierarchy.

Other

  • Docs updates

  • Several bugfixes and security patches.

v0.2.2

What’s new?

  • The MNIST examples (Keras and PyTorch) have been updated so that they now bundle the example data in .npz format.

Other

  • Docs updates

v0.2.1

What’s new?

  • It is now possible to choose which validation metrics to plot in the Dashboard

Fixes

  • Plots backed by no current data is no longer shown as empty plots.

Other

  • Docs updates

v0.2.0

What’s new?

  • Support for PyTorch models

  • It’s now possible to have examples in external repositories

  • Support for models constructed with the Keras Functional API

  • Set maximum number of clients in the settings file

  • Now possible to upload the compute package through the UI

  • Configure round timeout in the UI

combiner:
name:
combinerhost:
combinerport: 12080
max_clients: 50

  • Added several new views in UI including combiner performance and client training time distribution.

  • Updated look-and-feel for UI.

  • Added client allocation policy to spread the clients evenly over the combiners

  • Use config for s3 model commits instead of a hard-coded bucket name

  • Memory management to prevent combiners from going off

  • Reducer, client and combiner now have their own separate Dockerfile definitions

Fixes

  • Harmonizes naming of helper classes

  • Many bugfixes and stability improvements

Other

  • Lots of product documentation updates