GlaDOS (Portal 1 & 2, Valve, Ov2 super trained) 150 Epochs

[quick wigeon]

This is a long overdue post but anyway funny robot lady
https://huggingface.co/C0ttontheBunny/Portalmodels/resolve/main/GlaDOSOv2V2.zip?download=true
VA: Ellen McLain

[jolly harbor]

Rate the model

To rate this model, either use /rate or click the buttons below.

[fleet ocean]

im happy that i beat portal 1

[quick wigeon]

both Portal games are so good man

[cobalt meadow]

Do yourself a favour and play both.

[fleet ocean]

i have both anyways and im at the part in portal 2 where wheatley is in control.

[quick wigeon]

"I am a potato"

[fleet ocean]

Clap* Clap* Thank God Thats Still Working.

[vagrant steppe]

GLaDOS sounds like she's trapped me in a room and is trying to give me a lesson on lesbians

[quick wigeon]

This is what it feels like to be a personality core or a turret in Aperture

[vagrant steppe]

this ones the best one
a simple robo effect thingy would make it way better fr

[quick wigeon]

she has it if you speak like her in a monotone voice like the "s" and sometimes "f" have a robotic sound to them

[vagrant steppe]

ye but not every one is going to speak like that

or well ig if you OVERTRAIN it

to like FORCE it to be like the dataset

[quick wigeon]

true, I do it when using the model with real time just to make it actually sound like her

[quick wigeon]

I could never

[vagrant steppe]

Overtraining can both be good and bad

sometimes its good

[quick wigeon]

I guess depending on what your model is supposed to sound like

[vagrant steppe]

even if you want it to sound human
from my tom foolery
sometimes the over trained models sounds better tbh
But well it might as well NOT be over trainned cause im just basing this on the two graphs of total loss
even tho thats NOT how you tell if its over training or not

grad norm and then loss fm and loss mel and so many other stuff

so i just never bother with graphs and tell by just testing and picking the best one

[quick wigeon]

That's the best way to check anyway

[vagrant steppe]

true

its the easiest atleast

set up a W&B for RVC 😭 /j dont even bother

[quick wigeon]

A wha

[vagrant steppe]

Weights and Biases

its the holy of graphs

its a mess to set it up but its great

https://wandb.ai/site

literally try looking at a guide for this shit

"basic" set up https://docs.wandb.ai/guides/hosting/how-to-guides/basic-setup

also ye its useless to set this up for RVC
RVC is NOT deserving of this
and not worth it

[quick wigeon]

I honestly don't know too much about RVC or training ai models in general

[vagrant steppe]

same
you just do this and goofy parameters and go brrt

batch 4 for dataset of 10 min
and 8 for 20 min
we dont talk about the code level tweaks but https://github.com/Enrop/RVC-Edits

[quick wigeon]

Fr I always use batch 8 for everything because idk how it works

[vagrant steppe]

high batch size is learning fast and a low batch size is learning slower
so better for smaller datasets

dont do batch size 1

[quick wigeon]

Batch size 1 sounds like hell

How long for just one singular epoch?

[quick wigeon]

I had batch size 8 on a 9 second dataset of a lego tusken raider

I probably should redo it with size 4 or something

[vagrant steppe]

not only slow
but it will make your model shit tier

[vagrant steppe]

i'd do 3 tbh

[snow rune]

Model Ready ✅

This model has been synced with Weights and is ready to use for free!