v0.1.8

• Query Key Value version of cross attention layer
• add multi (homograph, homonym) trainer and dataset
  • fix bug: we were hashing the placeholder symbol _ as the solution word
  • Changes to multi transformer and to it's dataset
• implement masked attention but not use it (it performed worse)

What's Changed

• rename dirty to lexicon by @thewh1teagle in #2

New Contributors

• @thewh1teagle made their first contribution in #2

Full Changelog: v0.1.7...v0.1.8

v0.1.7

• add phonemizers (transformer) and their dataset
• add deps
• doc fix
• add sum and cross attention layer

Full Changelog: v0.1.6...v0.1.7

v0.1.6

• fix loading bug of certain models

Full Changelog: v0.1.5...v0.1.6

v0.1.5

• global premodulo support
• quaternary training support
• fix slight inneficiency

Full Changelog: v0.1.4...v0.1.5

v0.1.4

• fix cuda
• fix tiny bug in the new cpu algo
• make learning.Reducing() public

Full Changelog: v0.1.3...v0.1.4

v0.1.3

• cuda learning is broken in this release
• fix build on android/arm64

Full Changelog: v0.1.2...v0.1.3

v0.1.2

• new experimental learning
  • cuda learning is broken in this release, might be fixed later
• move the pre-xorring+subtracting burden to json loader/writer
• only build vectorized hash on supported archs (fixes goruut build)

Full Changelog: v0.1.1...v0.1.2

v0.1.1

• multiclassifier updates
• finalized phonemizer training script. This phonemizer trainer (v0.1.1) will be used for the next goruut release.
  • fix loading of zlib
  • used transition memoization optimalization, use improvement possible
  • remember local minima and avoid them
• parallel: add LoopUntil()

Full Changelog: v0.1.0...v0.1.1

v0.1.0

• Tally4 is now mandatory - battle tested on various datasets
• Infer2 is now mandatory
• Breaking change to majority pooling is recommended, use New2, New will be removed in future major version
• New pretrained models
• Parallel package with thread safe loop

Full Changelog: v0.0.12...v0.1.0

v0.0.12

• add pprofs
• Make Huge dataset less huge
• Updated commands: train_is_alnum, train_is_virus, train_squareroot
• use parallel loops
• Squareroot demo
• feedforward net: Tally4, Infer2, GetBits and GetClasses

Full Changelog: v0.0.11...v0.0.12