EveryVoiceTTS / EveryVoiceTTS/EveryVoice
Resuming training for text-to-spec model does not resume with appropriate losses
Open
bug
- Dominant language
- Python
- Stars
- 45
- Forks
- 4
- Avg merge
- 1d 2h
- Merged PRs (30d)
- 14
Description
To recreate: train a text-to-spec model, then add a `finetune_ckpt` value and resume training (you might need to change another model value like the max number of steps). the losses do not resume properly.
Maybe related to https://github.com/EveryVoiceTTS/EveryVoice/pull/473
#419
Contributor guide
Assessment
This issue has not been assessed yet.