Software Engineering Courses (SECourses)•13mo ago

for 3090 you need to use 24 gb configs

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

depending on how much you want to wait

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

each one has speed in the config name

Nneo hey dr @Furkan Gözükara SECourses

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

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

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

but i agree all gpus are out of stock

Furkan Gözükara SECoursesOP•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 SECoursesOP•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...

neo•12/14/24, 11:06 AM

what about vast.ai ?

Nneo what about vast.ai ?

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

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

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

just extra burden

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

i need massed compute alikes that gives us coupon

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

ye trying

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

made it work damn so nice

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

let me test

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

doing final reinstall haha

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

man it took me a while always a new problem

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

it doesnt sadly

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

but it is faster

DDazzastrous [4090]wolf howling at the moon

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

i would say test all cases and see how they perform

Furkan Gözükara SECoursesOP•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: