Doug Walker/Nostalgia Critic

np <:Joe_Epic94:1093358443198423050>

Whats the amount of minutes you had in your dataset?

over 20 minutes iirc

so around ~22

I recommend having it go up to around 750-1000

Check the Total/g/loss for when it starts going up near that point to make sure it doesn't overtrain, and you will have an improved model.

...minutes?

i'm going to have to get a lot of training data then lol

No Epoch

750-1000 epochs

ohhhh

well once i find a way to make google colab finally work again i'll be sure to do that

You have a good pc?

...define "good pc"

ig it's good for games but it doesn't seem to be the case for doing ai stuff like this

(talkin about a gtx 1650 here, i'm not getting anywhere with it; i'm slowly realizing that)

a 1650 is still good, but slow for local.

I have a gtx 1070 and its slow for training.

slow? man it doesn't even *work* for local

it's vram goes out of memory

really?

thats news to me.

it only ever works when i set the batch size to 1 lol

at that point it's a miracle that it can get even one epoch done

the best luck i've had is using it with the colab, and even then it's a nightmare

most of the time it can't connect to my gpu, but for some strange reason it just worked one time and then never again

and most of the time it spits out errors for literally every single step i go through in the gradio ui

and keep in mind that that happened to me for litereally every colab link i tried out <:nails:1091517541240807424>

i'm not exactly sure honestly what i'm doing wrong-- but i think it's just my pc that's just not built for this kind of stuff

I feel like my best bet would be just getting a much more powerful GPU that has 8 gigs of VRAM... otherwise by the time I'm done I'll be long dead