Tesla T4 (training on colab)

How's your config?

I'm using the notebook's defaults

Just needed to reduce the batch size so it doesn't OUM on me

Yeah but that's not an error with OUM

I train with T4 and batch 18 with no problems

The default batch size is 23 in the notebook...

Reducing helped but then this error spawned

After the OUM error?

If yes, try to restart runtime and try again

Lets try

Even after completely re-starting from scratch this CUDA error appears, huh

For now I'm killing the runtime, will get back to it later today

The error still seems to appear Since it's a repeating issue, it could be the PL notebook being a lil outdated More peeps would need to encounter it though just to ensure

What's OUM lol?

Out Ufff Memory?

Ye

But this one is easy to solve

The other error I'm occuring... Not so much

I'm trying to use denoiser with hifigan but it doesn't work, I don't get why, the code is the same as other synthesizing colabs I've seen, I don't know what I'm doing wrong... I've tried to remove the "device" part with no luck

I got it to make it work, I had to change uberduck_ml_dev/models/common.py for it to work

hmmmm

is this a new error?

happy to review a PR

No, it's okay, this is just a modification to use Super Resolution and Avocodo, I have a new error though, I've tried training using 2 gpu's but I wasn't able to make it work. I changed the backend to gloo and distributed_run to true, were you able to train using 2 or more gpu's using Windows?

i made a hifi gan for tacotron legacy and i spent 4000 steps aka a lot of hours on it and it doesn't work? anyone know why or?

It's probably that error when it starts at 250000 steps, I think you may have to re-train

I had this problem before

Which colab are you using?

i used the hifi gan training colab and CPU hifi gan synthesis both in #841437191073955920