how to train models on Termux?

[vast burrow]

who said that? Mobile GPUs wouldn't be enough powerful for that, and i dont think they would be compatible with cuda that its of Nvidia GPUs

[digital folio]

Okay thanks, but one of the staff/behind Weights
Already told me that we can't train locally on phone ☹️

[vast burrow]

yea

[elfin veldt]

at least you can inference in Termux?

[vast burrow]

Maybe with CPU but idk tbh

[elfin veldt]

Probably a 10% slower than a low end CPU if its not a Snapdragon 8+ gen 1 idk

[stoic flume]

https://tenor.com/view/misinformation-gif-26306186

[digital folio]

Yeah because we can't train models locally on mobile

[stoic flume]

do u think he has really tried training a model on it for several months?

[digital folio]

Maybe

[livid cairn]

https://github.com/wasdwasd0105/limbo_tensor

Hear me out

[digital folio]

what is this?

[elfin veldt]

according to what i read, it's a vm of windows/linux for google pixel phones

mostly google pixel 6, 7 and 8.

probably bc this vm is only for high end phones tbh idk why

[elfin veldt]

i thought that this was to run tensorboard on mobile 😭

[digital folio]

Yeah, but I'm Xiaomi Redmi user and idk if Xiaomi phones are consider as "high end phone" but it's cheaper than Google Pixel and I think only very expensive phones like iPhone,Google Pixels etc. are considering "high end phones"

[livid cairn]

I'll install applio and screenshot the inevitable error

[livid cairn]

Google phones are high end price for mid tier performance

I guess there's just some special locked away virtualization stuff that that thing unlocks

[elfin veldt]

u cant train on applio using a google pixel 6, 7 or 8 phone using termux

🧠

[livid cairn]

Python aint even gonna work on this hoe, its windows 11 on arm

like i might be able to make it work if i get the right builds n shit but im not down for all that

[elfin veldt]

😂

lol

in the first instance, what would happen is that your cell phone would burn out.

also Python does not work well on mobile

not even with the most powerful phone could you train

[elfin veldt]

https://github.com/IAHispano/Applio/tree/exp/cpu-training

[vast burrow]

thisturtlehasseensomeshit

[elfin veldt]

just applio

[vast burrow]

training on cpu would take soooo much time

[elfin veldt]

works

blaise pc lol

[vast burrow]

yes u could already do that with the 'fake gpu' which is basically cpu, its just gonna take sooooooo much time

[elfin veldt]

yea

[vast burrow]

thats why training on cpu is worthless

[elfin veldt]

I think Blaise will improve him to train faster.

[wraith delta]

Ayo? @elfin veldt level 17 !!!

[vast burrow]

he sent this in ai hispano?

[elfin veldt]

yes

[vast burrow]

cooking cpus fr

[elfin veldt]

spanish help chat

[elfin veldt]

they used to say that with training with amd gpus and now they train the same as an rtx.

[vast burrow]

because applio now has zluda support, which is cuda for amd basically

but cpus suck for training

[elfin veldt]

before that someone was able to train locally on a mac

[vast burrow]

rvc doesn't have mac support for training, only inference, so its the same as cpu training

where did you see someone train locally on mac btw?

[elfin veldt]

ai hispano

[vast burrow]

bte there is also Winlator

should be better, its used for gaming pc games usually

[vast burrow]

idk how the development is going, but usually no one trains any ai on cpu as its just slower when they can use cloud gpus for free, i doubt it will be as good as gpu

[digital folio]

Okay, but I will delete this my #1192011222023950368 post because I already know that we can't train models on Termux as a mobile user

[livid cairn]

Or you can never close the thread and let it devolve into a shitpost thread

[elfin veldt]

move it to the depths of ai hub

lol

[elfin veldt]

https://discord.com/channels/1159260121998827560/1282556233810706443

[livid cairn]

I offered some helpful insight in there

Very well written response I made

[vast burrow]

https://discord.com/channels/1159260121998827560/1289538710307602554

Well u can but need like a super gaming phone, i was able to only inference cus it crashed while training

[digital folio]

Yeah but I'm probably gonna buy laptop for that (also for Replay and unlimited GPU)

[elfin veldt]

red magic 9 pro

[stoic flume]

ROG Ally that runs windows at this point

[vast burrow]

its better a desktop pc with an rtx 3060 than a laptop as usually laptops have either integrated graphics or rtx 3050 which has low vram for anything

[vast burrow]

maybe

[vast burrow]

fr

[elfin veldt]

if NVIDIA adds cuda to snapdragon processors, there's a chance we can run ai stuff

[vast burrow]

I mean u can still run ai stuff on CPU, but i really doubt nvidia would do such a thing

maybe a 'zluda' version for android

[elfin veldt]

or maybe snapdragon may do a "cuda" itself exclusively for snapdragon processors

and we had to use virtual cuda just like "zluda"

[vast burrow]

maybe

Only AI local app i seen using gpu is https://github.com/tumuyan/RealSR-NCNN-Android

which is built with android studio

idk how exactly it works tho