[Question] 124GB Model from 8 minutes of audio #444

New Issue

Hitmare · 2023-11-09T16:12:34Z

Hitmare commented

2023-11-09 16:12:34 +00:00

HI

I'm new to the AI voice clone and didn't find any references and numbers how big a training/finetuning should be

I've trained a model based on 8-9 minutes of audio and the result in the ./training/ folder is around 124gb in size. is this something normal or did i miss something?

BR

HI I'm new to the AI voice clone and didn't find any references and numbers how big a training/finetuning should be I've trained a model based on 8-9 minutes of audio and the result in the ./training/ folder is around 124gb in size. is this something normal or did i miss something? BR

FortermalGreek commented

2023-11-09 18:09:17 +00:00

You probably have set your 'Save Frequency (in epochs)' to a very low number, meaning, every couple of epochs, a new model and training state is saved.

In the training/finetuning go to the "models" folder, you should see a lot of .pth files, remove all of them, but keep the one with the highest number in the name

Go back to training/finetuning and go to the "training_state" folder then do the same thing

The ones with the lower number are just less trained models.

You probably have set your 'Save Frequency (in epochs)' to a very low number, meaning, every couple of epochs, a new model and training state is saved. In the training/finetuning go to the "models" folder, you should see a lot of .pth files, remove all of them, but keep the one with the highest number in the name Go back to training/finetuning and go to the "training_state" folder then do the same thing The ones with the lower number are just less trained models.

👍 1

Sign in to join this conversation.