Skip to content

Latest commit

 

History

History
9 lines (6 loc) · 437 Bytes

roadmap.md

File metadata and controls

9 lines (6 loc) · 437 Bytes

Roadmap

For bugs and other issues see the repository issues on GitHub.

General Improvements

  • Load training data from a file instead of storing it in memory to allow for quick resume of training without needing to regenerate the dataset from the source corpus.

  • Read training data from a file in chunks to allow for datasets beyond the size of available RAM.

  • Output attention alignments correctly in tensorboard summary logs.