v0.3.0

What's Changed

• Fixes related to restoring models and logging by @eginhard and @AntoBrandi in #18:
  • Ensure step/epoch counts are restored correctly on continue, but reset to zero on restore.
  • Adopt the standard Pytorch scheduler interface for Coqui's custom LR schedulers.
  • Also store and restore the scheduler state_dict, but only restore optimizer and scheduler on continue.
  • Correctly display total number of steps/epochs because they are zero-indexed.
  • Log learning rates of the current step, not the next one.
• fix: remove use of Numpy in stepwise scheduler by @eginhard in #19

New Contributors

• @AntoBrandi made their first contribution in #18

Full Changelog: v0.2.3...v0.3.0