Can I use a recipe to fine tune a model ?

[Ca-ressemble-a-du-fake]

Hi,

I am trying to fine tune the Exp3 model (En-PT-Fr). So far I used the VITS recipe, customized it and launched the fine tuning from the downloaded checkpoint :

!python3 /content/drive/MyDrive/YourTTS/TTS/configs/vits_ljspeech.py \ --restore_path /content/drive/MyDrive/YourTTS/Models/Experiment3/best_model_latest.pth.tar
My vits_ljspeech.py has vitsargs (model args), datasets, ... set up,

But watching @thorstenMueller 's tutorial where he is using config_path path/to/config.json I now wonder if I am using the recipe the right way.

Is there any difference between launching the training from a recipe (ie a custom python script) and launching it from train_vits.py with custom config.json ? Which way should I choose to fine tune a model ?

Thanks

[erogol]

Doesn' t matter how you run the code. What matters is the config fields, arguments. You should also use the --restore_model flag to point to the model you want to fine-tune.

[Ca-ressemble-a-du-fake]

Thank you @erogol , I'll have a look at --restore_model flag!