mrq/ai-voice-cloning
31
74
Fork 43
Code Issues 223 Pull Requests 5 Packages Projects Releases Wiki Activity

Update 'Training'

mrq 2023-03-13 19:02:29 +00:00
parent f949055598
commit af30bfa560
1 changed files with 2 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Training.md

@ -118,12 +118,10 @@ After filling in the values, click `Save Training Configuration`, and it should
### Suggested Settings ### Suggested Settings
The following settings are robust enough that I can suggest them, for small or large datasets. The following settings are robust enough that I can suggest them, for small or large datasets.
* Epochs: `100` (50 is usually "enough", large datasets can get 20) * Epochs: `50` (increase/decrease as needed, according to how big your dataset is)
* Learning Rate: `0.0001` * Learning Rate: `0.0001`
* Learning Rate Scheme: `MultiStepLR` * Learning Rate Scheme: `MultiStepLR`
* Learning Rate Schedule: * Learning Rate Schedule: `[2, 4, 9, 18, 25, 33, 50, 59]`
- small datasets: `[9, 18, 25, 33, 50, 59]`
- large datasets: `[2, 4, 9, 18, 25, 33, 50, 59]`
However, if you want accuracy, I suggest an LR of 1e-5 (0.00001), as longer training at low LRs definitely make the best models. However, if you want accuracy, I suggest an LR of 1e-5 (0.00001), as longer training at low LRs definitely make the best models.