v0.7.1 - Aquarius 1

Aquarius is a stabilisation release.

We have been hard at work on improving a number of features across the whole of the ledger, etch and machine learning library. The following is a brief overview of the updates that have been incorporated into this release:

Core

• Improved main chain sync API
• Enhanced transaction status API
• Main chain sync time improvements
• YAML configuration parser added
• Additional telemetry support for constellation
• Internal updates to the VM to enable support for initialiser lists
• HTTP server authentication and bug fixes
• Various bug fixes including exception safety issues

Etch

• Persistent globals support
• UTF8 String and Comment Support
• State saving implemented for Graph, Dataloader, and Optimiser

Math Library

• N-dimensional implementations for softmax & reduce
• reworking Reduce/Apply methods for aligned and non-aligned SIMD
• AVX2 support for all types

ML Library

• Example implementation of BERT language model
• Example updates: semi-synchronous distributed learning
• Node/Op relationship reworked from inheritance to ownership
• weight sharing reworked as nodes sharing Ops
• Node caching improved
• Serialisers for Graph, Ops, and Layers
• Serialisers for SGD optimiser and Tensor Dataloader
• Estimator reworked as 'model'
• Dataloader train/test splitting functionality

Developer

• Link-time optimisation support added to build by default