Holy lord this is undertrained
#Post
keen timber
Only 100 epochs for 26 seconds...
@ionic dirge give me your opinion on this silly situation Mr worm
ionic dirge
needed more time
thin sedge
I only have these data and how many epochs should I train?
keen timber
We cant tell you how many, you needa learn how to detect overtraining
balmy schooner
Holy lord this is undertrained
ur wrong
balmy schooner
Mmmmm
26 seconds of data is nothing
everything is collapsing as soon he clicks start training
rvc is still able to reproduce stuff because it's forced to do it so, but the model inside is completely fried
the model "learned" the whole dataset in 1 epoch
by training more ur just forcing the ai to only be able to reproduce the dataset
an epoch is everytime ur model has seen it's full dataset
if u force ur model to see your 10 min set for 1000 epochs (random example), ur just telling the ai to memorize the dataset
keen timber
whats your point
balmy schooner
this model is not undertrained
keen timber
so according to your logic, if this model was at 50 epochs it wouldnt be undertrained
balmy schooner
yep
keen timber
Ok, now I dont believe your logic anymore
balmy schooner
it's basic ml knowledge but fine then
thin sedge
Ehh I retrained the model and I got this: Lowest generator loss: 4.576 at epoch 10, step 37
thin sedge
Tenna (Deltarune) 50 epochs
thin sedge
We cant tell you how many, you needa learn how to detect overtraining
I just learned now, thanks
ionic dirge
