v2.0.23

What's Changed

• Updated workflows to contain Emerge binary size detection
• Improved RNN / LSTM
• Bug fixes

• Merge develop by @wvabrinskas in #91

Full Changelog: 2.0.22...2.0.23

v2.0.22

What's Changed

• Fixed issues with RNN
• Added NetworkContext to forward passes
• Added GeLu layer
• Fixed calculation issue with LayerNorm
• Added symmetrical arithmetic to Tensor named applyAlong. You can now add single dimensional Tensors to N dimensional tensors along the axis of the single dimension. This happens automatically for you.

Full Changelog: 2.0.21...2.0.22

v2.0.21

What's Changed

• Merge develop by @wvabrinskas in #82

• updated to support Float16 on macOS universal

Full Changelog: 2.0.20...2.0.21

v2.0.20

• Fix a bug where more than 8 threads on optimizer caused a big delay

What's Changed

• Merge develop [Hotfix] by @wvabrinskas in #81

Full Changelog: 2.0.19...2.0.20

v2.0.19

What's Changed

• Genericized Tensor.Scalar to support Float and Float16
• Added weight decay option to Adam

• Develop by @wvabrinskas in #79

Full Changelog: 2.0.18...2.0.19

v2.0.18

What's Changed

• Fixed RNN and LSTM layers to be more responsive and accurate
• Fixed a variety of build warnings
• Added Contribution guidelines
• Added more documentation
• General bug fixes

• Merge Develop by @wvabrinskas in #76

Full Changelog: 2.0.17...2.0.18

v2.0.17

What's Changed

• Fix memory leak by @wvabrinskas in #68

Full Changelog: 2.0.16...2.0.17

v2.0.16

Significant performance improvements.

PSA: Run / Train using a RELEASE scheme. It is 10X faster

What's Changed

• Performance improvements by @wvabrinskas in #66

Full Changelog: 2.0.15...2.0.16

v2.0.15

What's Changed

• Update README.md by @wvabrinskas in #62
• Merge Develop by @wvabrinskas in #64

Full Changelog: 2.0.14...2.0.15

v2.0.14

What's Changed

• Fix dropout import by @wvabrinskas in #61

Full Changelog: 2.0.13...2.0.14