Betty 0.2.1

New release 0.2.1 for Betty.

• New examples: Robust NAS (DSRNA), NAS Augmented Image Captioning (IUC), Saliency-Aware NAS (SANAS), ImageNet Data Pruning.
• Add black formatting.
• Improve grad, param, epoch callback, configuration.
• Improve iMAML implementation.
• Distributed support for Env.
• Add SAMA.
• Add bfloat16 support, update precision interface.
• Various bug fixes and stability improvements.
• General code cleaning/updates.

Betty 0.2.0

New release 0.2.0 for Betty.

• Distributed training.
• ZeRO Optimizer.
• FSDP parallel training (experimental).
• General code cleaning/updates.

Betty 0.1.1

New release 0.1.1 for Betty.

Highlights

• General code cleanliness updates.
• Improve meta-weight net (MWN) implementation.
• Fix and improve wandb logger.
• Add some docstrings.
• Add MWN ITD-RMAD.
• Add higher-based and torchmeta-based IterativeProblem.
• Add dropout to MWN.
• Add gradient clipping for ImplicitProblem.
• Support multiple dataloaders.
• Add warmup option.
• Add configuration and warnings for Problem class.

Betty 0.1.0

Initial release 0.1.0 for Betty.

Highlights

• Implemented Problem and Engine classes as abstractions for multilevel optimization & generalized meta-learning.
• Initial documentation and API reference.
• Examples for HPO, NAS, data-reweighting, domain adaptation for pretraining & finetuning, implicit MAML.
• Training with gradient accumulation.
• FP16 training.
• Non-distributed data-parallel training.
• Initial suite of tests.