but I answered

it's completely certain that it won't

FUCK YOU REVERSE-CARDED ME

?

I mean the running for longer

past the current state

It made a checkpoint though

...and it's looking slightly closer to the pic you sent, but not by much

ok, kid

we do a little trolling

This is not off-topic

your not going to be given access to GPUs anymore if you keep the colab on for a while

I'm trying to clean out my drive to make room for more voices, but I keep getting the events.out files mixed up to the point where I don't know which character it belongs to. So I have some questions... 1, are they needed for the character to work on google colab? And 2, will deleting these screw the character up?

i usually dont have any luck with google colab in general, lol (and i have colab pro cuz the one thing i would use it for worked for me, but since then you can hardly get any of the "good" gpus on free version anymore) i tried like 2 different training notebooks and just had no luck with them whatsoever even though i did like every step

i guess my main problem with the first one that ive tried (and i think its the one you all use as well) is that the datasets have to be uploaded to the colab itself instead of just getting them from google drive... lets just say this is not very ideal if you have quite a lot of audio data and colab just doesn't like it, i guess

for the other one i tried i just get errors from installing the requirements it just fails to install them

you can change that behavior

"errors" huh?

luck has nothing to do with this

those are tensorboard "logfiles"

probably you won't need them

"There is a way to measure the acute emotional intelligence that has never gone out of style."

This sounds SO MUCH better oh my god

Entered one of the original prompts ("Heat doesn't bother good ol' Wasabi. He's hot stuff all the time..."), came out ALMOST EXACTLY the same

I'm training it a bit more. It crashed on 0.13(?) last night due to inactivity

I have increased the batch size from 4 to 8 in hopes that it works a bit better and is a bit faster and it doesn't OOM

Turns out it was actually a bit more than that. But now the loss is down to

0.152577

I will still train it for a little longer. I'm gonna leave it running while I'm at school

ok, I read the paper now

it seems to me like some very incremental progress over GlowTTS