mrq
/
ai-voice-cloning
31
73
Fork 42
Code Issues 221 Pull Requests 5 Packages Projects Releases Wiki Activity

Update 'Utilities'

master
mrq 2023-03-29 19:39:33 +07:00
parent 19624fa58c
commit c1d46199aa
1 changed files with 12 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
Utilities.md
Unescape Escape

@ -12,4 +12,15 @@ To import a voice, click `Import Voice`. Remember to click `Refresh Voice List`
### Tokenizer
This section is for debugging purposes while working with custom tokenizer vocabs. You can use this to see how a given input will tokenize.
This section is for debugging purposes while working with custom tokenizer vocabs. You can use this to see how a given input will tokenize.
### Model Mixing
Given `model A` and `model B`, you can merge two models together by a `weight` factor (0 will favor `model A`, 1 will favor `model B`). My very, very, impromptu and cursory test yielded already great results:
* given finetunes for Melina (Elden Ring) and Mitsuru (Persona 3) with an equal weight of 0.5:
- Mitsuru (Persona 3): https://vocaroo.com/16Fm7lzKFNoD
- Melina (Elden Ring): https://vocaroo.com/1gH7PxosRjUr
Mixing is pretty quick, and will output the new model at `./models/finetunes/modelA_modelB_weight_merge.pth`
Use cases for this will be noted at a later time.