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