Software Engineering Courses (SECourses)•13mo ago

hey dr @Furkan Gözükara SECourses

neoOP•12/14/24, 9:54 AM

do we got any alternatives like Massed Compute? other than runpod

neoOP•12/14/24, 9:54 AM

masses compute is too out of capacity these days.. can't count on it

neoOP•12/14/24, 9:54 AM

massed*

Pprotector131090 cogvideo is cool but Hunyuan is way better. Its sad that we need to wait for img...

Furkan Gözükara SECourses•12/14/24, 10:08 AM

and there is huge optimizations being made

Furkan Gözükara SECourses•12/14/24, 10:08 AM

kohya made it run on 12 gb

Original message was deleted

Furkan Gözükara SECourses•12/14/24, 10:09 AM

for 3090 you need to use 24 gb configs

Furkan Gözükara SECourses•12/14/24, 10:09 AM

depending on how much you want to wait

Furkan Gözükara SECourses•12/14/24, 10:09 AM

each one has speed in the config name

Nneo hey dr @Furkan Gözükara SECourses

Furkan Gözükara SECourses•12/14/24, 10:09 AM

there is 1 company messaged me but i didnt find time to invest resources on it yet

Furkan Gözükara SECourses•12/14/24, 10:10 AM

@NicB and @computersheep told me they are working ton increase the inventory

Furkan Gözükara SECourses•12/14/24, 10:10 AM

but i agree all gpus are out of stock

Furkan Gözükara SECourses•12/14/24, 11:06 AM

amazing : https://github.com/hkchengrex/MMAudio

GitHub

GitHub - hkchengrex/MMAudio: [arXiv 2024] Taming Multimodal Joint T...

[arXiv 2024] Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis - hkchengrex/MMAudio

Furkan Gözükara SECourses•12/14/24, 11:06 AM

i need to test

FFurkan Gözükara SECourses there is 1 company messaged me but i didnt find time to invest resources on it y...

neoOP•12/14/24, 11:06 AM

what about vast.ai ?

Nneo what about vast.ai ?

Furkan Gözükara SECourses•12/14/24, 11:09 AM

ye but while there is runpod i dont have any benefit of vast ai atm

Furkan Gözükara SECourses•12/14/24, 11:09 AM

just extra burden

Furkan Gözükara SECourses•12/14/24, 11:09 AM

i need massed compute alikes that gives us coupon

Furkan Gözükara SECourses•12/14/24, 11:09 AM

https://www.linkedin.com/posts/furkangozukara_as-some-of-you-know-i-am-doing-installers-activity-7273642073684082688-DuNa?utm_source=share&utm_medium=member_desktop

Furkan Gözükara on LinkedIn: As some of you know I am doing install...

As some of you know I am doing installers and tutorials for RunPod and Massed Compute which are cloud GPU services. Recently Massed Compute is always out of…

FFurkan Gözükara SECourses amazing : https://github.com/hkchengrex/MMAudio

Dazzastrous [4090]•12/14/24, 11:13 AM

I tried the demo on hugging face and was impressed

DDazzastrous [4090]I tried the demo on hugging face and was impressed

Furkan Gözükara SECourses•12/14/24, 11:19 AM

making installer atm

FFurkan Gözükara SECourses amazing : https://github.com/hkchengrex/MMAudio

TheToday•12/14/24, 12:14 PM

Wow

Furkan Gözükara SECourses•12/14/24, 12:38 PM

torch audio causing me problems trying to solve

FFurkan Gözükara SECourses for 3090 you need to use 24 gb configs

�

𝕏4C•12/14/24, 12:39 PM

Thank you

Dazzastrous [4090]•12/14/24, 12:39 PM

U can do it Doc....

DDazzastrous [4090]U can do it Doc....

Furkan Gözükara SECourses•12/14/24, 12:41 PM

ye trying

Furkan Gözükara SECourses•12/14/24, 1:49 PM

made it work damn so nice

Furkan Gözükara SECourses•12/14/24, 1:49 PM

prompt is just moon

Dazzastrous [4090]•12/14/24, 1:59 PM

wolf howling at the moon

DDazzastrous [4090]wolf howling at the moon

Furkan Gözükara SECourses•12/14/24, 2:08 PM

let me test

Furkan Gözükara SECourses•12/14/24, 2:08 PM

doing final reinstall haha

Furkan Gözükara SECourses•12/14/24, 2:08 PM

man it took me a while always a new problem

Furkan Gözükara SECourses•12/14/24, 2:09 PM

made it work without annoying conda

Timson•12/14/24, 2:21 PM

Has anyone compared training speed on A6000 vs L40S GPUs for FLUX Dreambooth fine tuning? Is L40S worth double price?

TTimson Has anyone compared training speed on A6000 vs L40S GPUs for FLUX Dreambooth fin...

Furkan Gözükara SECourses•12/14/24, 2:26 PM

it doesnt sadly

Furkan Gözükara SECourses•12/14/24, 2:26 PM

but it is faster

DDazzastrous [4090]wolf howling at the moon

Furkan Gözükara SECourses•12/14/24, 2:27 PM

Timson•12/14/24, 2:49 PM

Another noob question about training datasets: if I have mixed quality dataset, i.e. I have high quality images with little variation and I want to include other images with lower quality but higher variation (different emotions, clothing and backgrounds) should I upscale all images to the same resolution and aspect ratio or not? Or it will not make any difference?

Timson•12/14/24, 2:52 PM

I've noticed earlier there is huge influence of aspect ratio on inference in FLUX dev, such as 3:2 horizontal would give you professional photo incline, 16:9 - movie/series/ film grain, 2:1 vertical- phone / social network alike etc

Timson•12/14/24, 2:53 PM

It's even more prominent than direct prompt

TTimson I've noticed earlier there is huge influence of aspect ratio on inference in FLU...

Timson•12/14/24, 3:01 PM

this also got me thinking that comparing checkpoints with only single aspect ratio output somewhat limits it's scope, but not sure if making a grid with randomised AR is possible

TTimson Another noob question about training datasets: if I have mixed quality dataset, ...

Furkan Gözükara SECourses•12/14/24, 3:17 PM

i would say test all cases and see how they perform

Furkan Gözükara SECourses•12/14/24, 3:17 PM

hard to tell

FFurkan Gözükara SECourses i would say test all cases and see how they perform

Timson•12/14/24, 3:21 PM

but there is no fundamental problem on training with mixed images sizes in a set?

Dazzastrous [4090]•12/14/24, 4:21 PM

Could have used your airship Doc from Cog

cyberbol•12/14/24, 4:35 PM

What is it new option ReActor ?

Timson•12/14/24, 6:14 PM

Is it normal that saving flux dev fp16 checkpoint takes 6 minutes?
I've tested the drive on that remote to have above 500MB/s writing speed, I don't see any CPU or GPU load during this operation, what could be the bottleneck?

Timson•12/14/24, 6:14 PM

what is the "normal" duration of this operation?

Latentnaut•12/14/24, 6:35 PM

Hello group, do you think a finetune training of this product will capture all these little letters and details?

TTimson Is it normal that saving flux dev fp16 checkpoint takes 6 minutes? I've tested t...

Timson•12/14/24, 6:45 PM

figured this out, I cheeped out on 16GB RAM + 16 GB swap that gets 99% filled during checkpoint save. Lesson learned, seems you need 32GB RAM for seamless training :heher:

hey dr @Furkan Gözükara SECourses

Similar Threads

Similar Threads

Similar Threads