Expensive time to do finetune (HELP)

[fliperaa]

sorry anyone can help me i run finetune in my PC with an 9minute good quality audio, but in the first step "generating the dataset" it's locked in 75%, was runned 36986.8s and no complete.

my pc is an
r5 3600
rtx2060
16gb ram

i understand my pc maybe is limited, but this is a very long run time no ?

i used whisper large model

when i start finetune, receive this msg to att gradio:
IMPORTANT: You are using gradio version 4.32.2, however version 4.44.1 is available, please upgrade.

Really need upgrade ?

[erew123]

Hi @fliperaa The gradio version is a nothing, so dont worry about that.

As for the getting stuck at 75% it could be an out of memory issue OR a corrupt file.

I cant speak as to your dataset/audio files, however what I can offer is a file to test with.

https://filetransfer.io/data-package/woURvHoB

If you want to put this and ONLY this file in your put-voice-samples-in-here folder and test generating your dataset with a new project name:

I cant say exactly how long this would take on your hardware, but on a Nvidia 4070, its about a 2-5 minute task for it to break that file up, so I would image 4-7 minutes at a guess (on your hardware).

At least if you try this out, you can identify is it a hardware issue (memory etc) or is it (probably) a audio samples/dataset issue.

Thanks

[fliperaa]

@erew123 I formatted my computer completely and reinstalled the updated alltalks, now I can create my dataset perfectly in finetune, but when I get to training I get this message here: what do I do?

[erew123]

@fliperaa when did you last update alltalk? I posted an update about 5 hours ago to fine-tune. There should be no bugs, but lmk. If not, please git pull the most recent update.

[fliperaa]

Oh nice,I hadn't seen the new update, I'll update it here and test it, thanks!

[erew123]

@fliperaa can confirm current build is working fine w/training and logging metrics too:

As mentioned this has been a huge re-write part of which is documentation, so you should find most answers to any questions within the built in help or in the WIKI sections https://github.com/erew123/alltalk_tts/wiki#-xtts-finetuning-guides should you need more detail.

Thanks

[fliperaa]

@erew123 thanks bro, now it's working 100% correct!