Training my voice model error | AI HUB | Page 1

sly moss Sep 15, 2025, 9:47 PM

#

GPU: AMD 7900xtx
OS: Windows 11
Guide Followed: https://docs.aihub.gg/rvc/resources/training/

I got my data set ready it's 3 mins long but when I try to train a model it doesn't go through the entire data set, and gives me errors. I'm using Applio on local

Training

Last update: May 5, 2025

sly moss Sep 15, 2025, 10:07 PM

#

In my TensorBoard there's no Scalar tab on the side

lavish lagoon Sep 15, 2025, 10:40 PM

#

run-applio-amd.bat

#

set HIP_VISIBLE_DEVICES="0"

#

change it to a different device ID

#

if 0 does not work, change to 1

#

check training screen/advanced settings at the top so it shows your 7900

#

instead of AMD Radeon(tm)

#

if you're using Adrenalin driver 25.5.1+ you need HIP SDK 6.2.4+

#

📎 hip_sdk_624.txt

sly moss Sep 16, 2025, 12:41 AM

#

lavish lagoon set HIP_VISIBLE_DEVICES="0"

I've pressed enter and refresh

#

#

Also tried 0 or 1 here didn't work

lavish lagoon Sep 16, 2025, 12:48 AM

#

edit the .bat file

#

not the UI

#

change "0" to "1"

#

restart applio

sly moss Sep 16, 2025, 12:48 AM

#

I tried it was still the same

lavish lagoon Sep 16, 2025, 12:49 AM

#

that's not possible

sly moss Sep 16, 2025, 12:53 AM

#

Yeah here, I don't know why

lavish lagoon Sep 16, 2025, 12:56 AM

#

I told you to change "0" to "1"

#

why did you not?

sly moss Sep 16, 2025, 12:57 AM

#

In the video I showed 0 and 1

lavish lagoon Sep 16, 2025, 12:57 AM

#

you did not save the file

#

you just closed the notepad

sly moss Sep 16, 2025, 12:59 AM

#

lavish lagoon Sep 16, 2025, 1:00 AM

#

task manager/performance screenshot

#

#

print(torch.cuda.get_device_name(0))

sly moss Sep 16, 2025, 1:03 AM

#

My 7900 is at 1

lavish lagoon Sep 16, 2025, 1:05 AM

#

do you actually use iGPU?

sly moss Sep 16, 2025, 1:06 AM

#

For my secondary monitor

lavish lagoon Sep 16, 2025, 1:06 AM

#

you can just connect it to 7900

sly moss Sep 16, 2025, 1:06 AM

#

Okay, I'll try that

lavish lagoon Sep 16, 2025, 1:06 AM

#

HIP_VISIBLE_DEVICES should work regardless, not sure why it does not

#

alternatively if you remove that set line you should be able to pick the right device on traning screen (0 or 1)

sly moss Sep 16, 2025, 1:15 AM

#

Yeah it just likes the number 0 ig

lavish lagoon Sep 16, 2025, 1:16 AM

#

well, with igpu unused it is not longer available, so normal gpu goes to 0

sly moss Sep 16, 2025, 1:17 AM

#

Okay that makes sense

#

So okay, the other one is there a reason why my tensorboard doesn't have the scalars tab?

#

Do I have to start training before it shows up?

lavish lagoon Sep 16, 2025, 1:20 AM

#

you can click refresh once the training actually starts logging

sly moss Sep 16, 2025, 1:25 AM

#

I'll get to the point like in the screenshot and see if I still get the same errors, brb

#

So extracting features seem to work this time, but is it normal for the preprocess dataset to only read 3 seconds of audio? Even if the prepared dataset is 3 mins long?

#

lavish lagoon Sep 16, 2025, 1:38 AM

#

the compilation may take a while

sly moss Sep 16, 2025, 1:40 AM

#

Yeah, but the preprocess stage was completed in 6 secs on 3 secs of audio

#

I'm fine with the wait

#

I'm just confuse why the preprocessing only looked at 3 secs of audio

lavish lagoon Sep 16, 2025, 1:41 AM

#

3 minutes

sly moss Sep 16, 2025, 1:42 AM

#

Oh

#

I thought that meant 3 seconds and 6 ms

#

That makes more sense now

#Training my voice model error