v0.3.2

Features

• Migrate to new torchrua

v0.3.1

Bug Fixes

• Resolve Meter.__cmp__ functions bug
• Resolve flush bug
• Resolve SchedulerMixin bug

v0.3.0

Features

• Add constant_ and orthogonal_
• Add Linear and ConjugatedLinear
• Add Projector and ConjugatedProjector
• Add optimizer Aku interfaces, i.e., SGD and Adam
• Add scheduler Aku interfaces, i.e., InverseDecay and Fixed
• Add GTU activation function
• Add SizedBatchSampler to support customized batch size
• Add meters, i.e., AverageMeter, AccuracyMeter, ClassificationMeter, and TimeMeter
• Add hooks, i.e., summary, load_checkpoint, save_checkpoint
• Add naive pipe for contextual word representation, i.e., CtxPipe

v0.2.1

Bug Fixes

• Fix trivial bugs of transformer.py
• Resolve vocab size bug

Documentation

• Add shape annotations

Features

• Add MultiHeadAttention
• Add TransformerFfn
• Add TransformerEncoderLayer
• Add TransformerDecoderLayer
• Add Transformer
• Update initialization functions
• Add padding_mask
• Add PosEmbedding and TriangularPosEmbedding
• Add IWSLT14 dataset
• Add PaddedListTensorPipe, PackedListTensorPipe, and CattedListTensorPipe
• Move PaddedListTensorPipe to padding.py
• Move PackedListTensorPipe to packing.py
• Remove original format.py

v0.2.0

Features

• Simplified most of Proc and Pipe with TorchRua v0.3.0

v0.1.2

Fixed some trivial bugs

v0.1.1

Bugfix and small updates

v0.1.0

Release prototype