Hmmm

I can't really talk much rn bc I am currently busy with college stuff... I also forgotten a bit of the process since it was about last month when I did it so...

I wonder why I have so many issues non stop

oh ok

Literally me during the 2-3 days of struggle trying to figure out how to train Tacotron 2 on my PC

However, once I start training again, maybe I can run through of how I trained on my PC, just can't do it today cuz it will a bit of a long guide

What kind of issues did u have?

Alright

I'm gonna set up kubuntu tri-boot anyway incase its a problem with stuff I have installed, and this is much easier on linux for sure

1st it's trying to get it to work, not utilising the GPU for some reason... 2nd, was the repositories installed being wrong or can't install, so I have to find alternatives... 3rd, finally got to train, but the speaker was TERRIBLE at coherent words, afterwards referred to Cookie's formula and I'm good

how did you solve not utilising the gpu? thats my issue on windows

also whats cookies formula?

Basically, I was installing the wrong version of CUDA Toolkit, so, I installed the right version and Pytorch finally worked and installed properly on me

Cookie's learning rate formula so that the speaker will say accurate and coherent words

on windows my problem is that it works but it just uses my cpu

link doesn't work on mobile ill have to look later

Huh... then you got to set the virtual enviroment for using tensorflow-gpu

I think...

yep its using that

tensorflow normal isn't installed at all

I did tensorflow in a script to list devices and the gpu is there

it just uses 0% gpu for some reason

and nvidia-smi doesn't show it even using it

Well, in that case... I don't really know cuz I don't remember something like that happening to me

yeah that's why I want ti fresh install ubuntu

my validation loss isn't decreasing and it looks way too early for it to happen. is this normal? any tips as to what i did wrong? i've got 191 samples ranging from 2-10 seconds and they're good quality imo, using the TalkNet notebook

it should be good enough

it will use the latest one anyway