Local Applio using CPU for training model

[lean sapphire]

Hello, I just started training a model with RVC2 Applio with my 13900K RTX 4090 computer. However, it is using my CPU, not my GPU? Is this normal? I thought it was meant to be using my VRAM.

[unkempt maple]

which Applio version? its runtime packages are supposed to have included PyTorch+cuda (2.1.1+cu121, but not sure on the latest applio), but if you still have the issue, try update the Nvidia driver and install CUDA toolkit (if needed?)

[lean sapphire]

applio 3.2.0

[silk perch]

Ayo? @lean sapphire level 1 !!!

[lean sapphire]

I restarted applio completely and started training from the beginning and now its working for some reason, very weird

my gpu has 24gbs vram but before i set batchsize 4 due to that being more accurate

[unkempt maple]

are you working on some 48k studio session? if not, you shouldn't use that sample rate

[lean sapphire]

I chose it because the file says 48khz in mediainfo

[unkempt maple]

its format isn't relevant, if it comes from any lossy sources like youtube, etc, its spectrogram cutoff will always suggest either 32k or 40k (in case of this example, it's 32k)

[lean sapphire]

ok thank you, the voice im trying to train has a radio filter and is a character from an old game so it would be lossy

[lean sapphire]

1 big thing i dont understand is how long the voice data set can be for even more quality, i have a lot of dialogue