#error while training
1 messages · Page 1 of 1 (latest)
you know who else is doing wrong , file was converted? MY MOM!
you need to link to a folder which has audio only, so extract that folder
its only audio in that .zip folder though
yea but it does need to be extracted
so you can put the path as a folder only
RVC stuff 
very.. very confusing
let me extract it
ah it worked, do you mind helping me set the rest of it up?
im braindead currently i got no sleep
sure sure, i can also link a guide for you too and then you can tell me if something goes wrong if you prefer that
I think i tried that, i didnt really understand it at all, im pretty much almost done so it wont take long
Ayo? @dull mortar level 3 !!! 
I managed to get here, what do i proceed with
should be fine as is
just feature extract
oh how come
rvmpe_gpu ftw
has better quality than harvest, faster, and basically has the same positive aspects as harvest
I already feature extracted it but i did it again on that mode, did i mess anything ip?
ah, the website lied lol
mmmm I don't think so
train feature index first
i just click that?
so we can see if anything screwed up
nothing else?
(2900, 768),74
training
adding
成功构建索引,added_IVF74_Flat_nprobe_1_tris_v2.index
alrighty, nothing bad happened
now we're gonna set up the training par
t
save frequency, set it to 10
total epochs, keep as is
batch size depends on the size of your dataset, I'd probably go for 8 for a 10+ min dataset
Save only the latest ckpt - yes
Cache training set - no (unless you have a smaller dataset)
Save a small finished model - yes
INFO:tris:Saving model and optimizer state at epoch 20 to ./logs\tris\G_2333333.pth
INFO:tris:Saving model and optimizer state at epoch 20 to ./logs\tris\D_2333333.pth
INFO:tris:saving ckpt tris_e20:Success.
INFO:tris:====> Epoch: 20 [2024-05-31 09:50:59] | (0:00:03.510162)
INFO:tris:Training is done. The program is closed.
INFO:tris:saving final ckpt:Success.
it should update you on how many epochs you're training it to
uhm
that's
a little bit of a low epoch count
i have no idea what any of it means
nvm I screwed up and couldn't read 💀
set total epochs to like 200, and train until then
this?
set that to 200, train again
assets, weights
training cycle. every time it trains your entire dataset it's an epoch
ohh so the more the better?
each and every epoch makes it closer to your dataset, but also there can be overtraining
I see
though overtraining takes... a while, at least
yeah its at 116
at 200 epochs you should have a decent model i suppose
what if it was 500
alright its done
not sure which one it is though 😭
you would know that, i have no idea what any of this is
like literally, i know NOTHING
you should know how big your dataset is though
it goes up to 1k
the audios, that's what i mean
the audios are like 30 seconds to a minute long
is that bad?
do i need to shorten the audios
mmm
small, but it could work
now you can just test the model
which .pth do i use
latest one, so, 200 epoch
dont i also need another thing other than pth
index isn't necessary but
eh, better to have it than not have it
it should be in the index dropdown as well
Ayo? @dull mortar level 4 !!! 
its not
was talking about inside the GUI, the index is inside logs/tris (aka your model's name)
also sorry, had to delete that vid cuz bg music was prob copyrighted
almost like I can get him into legal trouble
im kidding
but thank you for your help
i wouldnt struggled and got lost instantly without you lol
you're welcome! :)