CHANGELOG.md

Change Log

Unreleased

Breaking change!

New features

Tasks completed

Bug fixes

v0.5.1

Tasks completed

• Fix Numpy 2 compatibility (#137).
• Update conversion between Renyi-DP and approximate-DP based on CKS20 (#138).

v0.5.0

Breaking change!

• Removed Horovod support (#130).
• Upgrading minimum dependencies in docs and dev install-extras (#130).

New features

Tasks completed

• Support up to torch==2.6.0 and tensorflow==2.17.0 (#130).

Bug fixes

• FederatedAlgorithm should restore current iteration if crash during gather_results (#132).
• Fix xgboost model loading (#131).

v0.4.0

Breaking change!

• Callbacks have been split up into sub-modules under the module pfl.callback (#124).

Tasks completed

• Added packaging as a dependency, which was missing (#121).

v0.3.1

Tasks completed

• Allow Saveable to checkpoint itself on-demand (#115).

v0.3.0

New features

• Implemented MLXModel to support training with MLX (#80).
• Implemented Horovod-compatible Barrier class (#100).
• Implemented JointMechanism to enable applying different DP noise on subsets of the statistics (#105).
• Implemented JointPrivacyAccountant for joint privacy accounting in the case of using JointMechanism (#111).
• Implemented policy-based model checkpointing callbacks (#106).

Tasks completed

• Added MLX image classification example (#80).
• Added MLX language model example (#82).
• Added notebook tutorial for training with MLX (#81).
• Updated notebooks to work on Colab (#95).
• Added CITATION.cff (#99).
• Don't hardcode for CUDA 11.8 (#108).

Bug fixes

• Fixed bug in PyTorchFederatedDataset where it sometimes hang (#98).
• Fix PyTorchSeedScope for non-CPU random states (#100).
• Respect CUDA_VISIBLE_DEVICES if it is set for process (#100).
• Fixed algorithm states not working properly when restored from Saveable (#103).
• Fixed edge case for having HyperParams inside HyperParams (#112).

v0.2.0

Breaking change!

• EMMGMMHyperParams is renamed to EMGMMHyperParams (#55)

New features

• Return local metadata from model training to algorithm (#71).

Tasks completed

• Update FLAIR preprocessing script to download dataset from HuggingFace, available at https://huggingface.co/datasets/apple/flair (#72).
• Update LLM Benchmark Configs (#63).
• New improved worker scheduling in distributed simulations. Speeds up FLAIR benchmark by 19% (#73).
• Don't pin PyTorch version to 2.0.1 (#69).
• Move --noise_cohort_size to add_mechanism_arguments (#70).

Bug fixes

v0.1.0

2024-03-01

• Initial release!