Software Engineering Courses (SECourses)•10mo ago

Hey everyone, I have dual RTX 3090 linked with NVLINK, I've tried several ways to train lora using m

Hey everyone, I have dual RTX 3090 linked with NVLINK, I've tried several ways to train lora using multi GPUs but I always fail & get errors, I'm on windows & I tried Kohya SS & Ai-toolkit with and without accelerate launch with no luck, I'm not a developer & I don't know why I get an error everytime

So if anyone can help me using the installation

Zono50•3/28/25, 12:00 PM

@Furkan Gözükara SECourses how do i run cog2vlm in linux and have it use fp8?

Inoshi•3/28/25, 12:19 PM

https://github.com/FoundationVision/VAR

GitHub

GitHub - FoundationVision/VAR: [NeurIPS 2024 Best Paper][GPT beats ...

[NeurIPS 2024 Best Paper][GPT beats diffusion

] [scaling laws in visual generation

] Official impl. of "Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction&...

GitHub - FoundationVision/VAR: [NeurIPS 2024 Best Paper][GPT beats ...

IInoshi https://github.com/FoundationVision/VAR

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•3/28/25, 12:25 PM

How do you find these repositories?

Inoshi•3/28/25, 1:04 PM

i am into ai but also people i follow post these regularly, and i share those i believe can have an impact based on the recent trend (4o)to places i believe someone might be interested

⚔⚔Seifme⚔Hey everyone, I have dual RTX 3090 linked with NVLINK, I've tried several ways t...

Furkan Gözükara SECourses•3/28/25, 3:20 PM

multi gpu only works on linux as far as i know. also normalyl we use block swapping and it wasnt working with multi gpu for flux

ZZono50 <@205854764540362752> how do i run cog2vlm in linux and have it use fp8?

Furkan Gözükara SECourses•3/28/25, 3:20 PM

you can use runpod scripts

IInoshi https://github.com/FoundationVision/VAR

Furkan Gözükara SECourses•3/28/25, 3:20 PM

this is source of new gpt 4o model

Furkan Gözükara SECourses•3/28/25, 3:20 PM

but it is weak

FFurkan Gözükara SECourses this is source of new gpt 4o model

Inoshi•3/28/25, 3:42 PM

also this https://chat.inceptionlabs.ai/ do you think we will see bigger models using diffusion ?

IInoshi also this https://chat.inceptionlabs.ai/ do you think we will see bigger models ...

Furkan Gözükara SECourses•3/28/25, 11:11 PM

i dont think but we will see

Dazzastrous [4090]•3/29/25, 11:08 AM

heavily censored

Dazzastrous [4090]•3/29/25, 11:08 AM

stifles creativity

Dazzastrous [4090]•3/29/25, 11:09 AM

right lets have a look at wan, What version we upto Doc? starting to look like a airplanes cockpit with all the extra features

DDazzastrous [4090]right lets have a look at wan, What version we upto Doc? starting to look like ...

Furkan Gözükara SECourses•3/29/25, 12:14 PM

i got few more features in my mind hopefully will code asap

Dazzastrous [4090]•3/29/25, 2:25 PM

sorry but these versions of Wan are slowwwww even with 20 steps and fast presets taking forever

Dazzastrous [4090]•3/29/25, 2:28 PM

356.69 it/s

Dazzastrous [4090]•3/29/25, 2:28 PM

lol

Dazzastrous [4090]•3/29/25, 2:28 PM

only using the 480 13b model

DDazzastrous [4090]only using the 480 13b model

Furkan Gözükara SECourses•3/29/25, 2:32 PM

on rtx 3090 ti?

Furkan Gözükara SECourses•3/29/25, 2:37 PM

probably you are using too much vram

Furkan Gözükara SECourses•3/29/25, 2:37 PM

reduce number of parameters to 7b

Furkan Gözükara SECourses•3/29/25, 2:37 PM

my initial speed it will take probably like 10 min

Furkan Gözükara SECourses•3/29/25, 2:37 PM

Dazzastrous [4090]•3/29/25, 2:39 PM

on 4090

DDazzastrous [4090]on 4090

Furkan Gözükara SECourses•3/29/25, 2:40 PM

lol it is even faster

Dazzastrous [4090]•3/29/25, 2:40 PM

yeh i think thats the only paramter i didnt change

DDazzastrous [4090]yeh i think thats the only paramter i didnt change

Furkan Gözükara SECourses•3/29/25, 2:40 PM

i just updated app for 7b

Furkan Gözükara SECourses•3/29/25, 2:41 PM

i guess 8b is being edge case

Dazzastrous [4090]•3/29/25, 2:41 PM

Im sure it was faster in earlier versions though Doc?

Dazzastrous [4090]•3/29/25, 2:41 PM

dunno

DDazzastrous [4090]Im sure it was faster in earlier versions though Doc?

Furkan Gözükara SECourses•3/29/25, 2:41 PM

no last version way faster

Furkan Gözükara SECourses•3/29/25, 2:41 PM

you need to see power like this

Furkan Gözükara SECourses•3/29/25, 2:41 PM

Furkan Gözükara SECourses•3/29/25, 2:41 PM

or that means it is using shared vram

Dazzastrous [4090]•3/29/25, 2:41 PM

yehg its always sharing vram

DDazzastrous [4090]yehg its always sharing vram

Furkan Gözükara SECourses•3/29/25, 2:42 PM

Furkan Gözükara SECourses•3/29/25, 2:42 PM

8b using 22.7 GB VRAM

Dazzastrous [4090]•3/29/25, 2:43 PM

I remember making 12 secs vids lol

Dazzastrous [4090]•3/29/25, 2:43 PM

reducing fps to 8 and x4 R something

Dazzastrous [4090]•3/29/25, 2:44 PM

selecting 24 gb is ok yeh?

Dazzastrous [4090]•3/29/25, 2:44 PM

or should i switch to 16gb

Dazzastrous [4090]•3/29/25, 2:46 PM

what effect does using loras have on speed?

DDazzastrous [4090]what effect does using loras have on speed?

Furkan Gözükara SECourses•3/29/25, 2:50 PM

Furkan Gözükara SECourses•3/29/25, 2:50 PM

3090 ti - 480p image to video 81 frames

Furkan Gözükara SECourses•3/29/25, 2:50 PM

make this 7b : Number of Persistent Parameters In Dit (VRAM)

Weatherby•3/29/25, 3:14 PM

On my undervolted 3090 ti, wan img2vid 480p is running about 25 minutes or less atm.

Weatherby•3/29/25, 3:14 PM

Oh wait, scratch that!

Weatherby•3/29/25, 3:14 PM

On my 4060ti 16gb it's that. I haven't run it with the new optimizations on the 3090ti

Dazzastrous [4090]•3/29/25, 5:09 PM

I give up be here till next christmas i will try pinokios version seems to be faster mines still stalling even on the easy settings 22.4gb and 3.2 ram

Hey everyone, I have dual RTX 3090 linked with NVLINK, I've tried several ways to train lora using m

Similar Threads

Similar Threads

Similar Threads