Software Engineering Courses (SECourses)•13mo ago

I am curious to maximise the PCs ability - with 6x 4090 and 512 DDR5 RAM (PCIe 5) - to train models

I am curious to maximise the PCs ability - with 6x 4090 and 512 DDR5 RAM (PCIe 5) - to train models faster

FFurkan Gözükara SECourses you can do multi step to get what you want 🙂

Sipriyani•12/15/24, 1:57 AM

doctor what exactly you mean with multi step?

Sipriyani•12/15/24, 1:59 AM

also another question, can I re-finetune, a flux model file which is already fine tuned?

dsienra•12/15/24, 3:54 AM

https://civitai.com/models/1034630/hunyuan-video-arnold-schwarzenegger-lora?modelVersionId=1160430 Great hunyuan lora trained only with 15 images, this video model generalize very well. very promising

QQM I am curious to maximise the PCs ability - with 6x 4090 and 512 DDR5 RAM (PCIe 5...

XCogni•12/15/24, 10:16 AM

sell the 6 4090s, buy 2 6000adas, or wait for the new blackwell chips coming out

FFurkan Gözükara SECourses like few seconds

Timson•12/15/24, 11:24 AM

and what is you time for swapping fine-tuned full checkpoints in SwarmUI? Mine like 120s per switch

There is this lines in debug log during model switch are they ok? clip_l is loaded and working during generation, and it says something about casting although I'm saving checkpoints as fp16

FFurkan Gözükara SECourses published : https://www.patreon.com/posts/117990364

J2_Nedo•12/15/24, 12:14 PM

thanx this is really good for our own made videos

QQM I am curious to maximise the PCs ability - with 6x 4090 and 512 DDR5 RAM (PCIe 5...

Furkan Gözükara SECourses•12/15/24, 2:27 PM

maybe multi gpu lora but i am not much hopeful

SSipriyani doctor what exactly you mean with multi step?

Furkan Gözükara SECourses•12/15/24, 2:28 PM

like train 1 time generate pictures pick good ones train another time

SSipriyani also another question, can I re-finetune, a flux model file which is already fin...

Furkan Gözükara SECourses•12/15/24, 2:28 PM

i wouldnt recommend

Furkan Gözükara SECourses•12/15/24, 2:28 PM

flux already very prone to overfit

Renzo•12/15/24, 2:51 PM

@Dr. Furkan Gözükara, in which cases do you use YOLOv9? I have observed that you add it to some prompts and I wonder what makes you decide to add it.

RRenzo @Dr. Furkan Gözükara, in which cases do you use YOLOv9? I have observed that you...

Timson•12/15/24, 2:53 PM

the cases where character face is likely to be small in the image composition

Timson•12/15/24, 2:57 PM

I'm experimenting with fine-tuned models and so far I'm not entirly convinced they are better than loras in all cases, especially without additional inpainting techniques.
Maybe they are still under-trained, but I tested 1800, 2000, 2500 training steps on different training images set sizes (17, 25, 40) with mixed results, some seeds yield good likeness but some miss completely with the same prompts.

joaomkx12•12/15/24, 3:48 PM

Does anyone know good VAE and text encoders for SDXL and SD models?

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•12/15/24, 3:56 PM

How can I generate a grid in swarm ui with 7 Loras each lora generating 4 images?

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 How can I generate a grid in swarm ui with 7 Loras each lora generating 4 images...

Timson•12/15/24, 4:51 PM

just fill in loras in grid tool and eigher 4 prompts or a single prompt and 4 seeds

TTimson just fill in loras in grid tool and eigher 4 prompts or a single prompt and 4 se...

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•12/15/24, 5:07 PM

Thanks will give this a try

RRenzo @Dr. Furkan Gözükara, in which cases do you use YOLOv9? I have observed that you...

Furkan Gözükara SECourses•12/15/24, 9:05 PM

especially more distant shots

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 How can I generate a grid in swarm ui with 7 Loras each lora generating 4 images...

Furkan Gözükara SECourses•12/15/24, 9:17 PM

use lora grid tab

Furkan Gözükara SECourses•12/15/24, 9:17 PM

and seed

Furkan Gözükara SECourses•12/15/24, 9:37 PM

something huge coming

Furkan Gözükara SECourses•12/15/24, 9:37 PM

training flux with using redux to improve quality

TheToday•12/15/24, 10:17 PM

Does anyone know if massedcompute has monthly plans?

TTheToday Does anyone know if massedcompute has monthly plans?

Furkan Gözükara SECourses•12/15/24, 10:34 PM

@NicB @computersheep

Drocsid Imperator Maximus•12/15/24, 11:06 PM

Literally just recharged credits on MassedCompute, then this happens!

Drocsid Imperator Maximus•12/15/24, 11:08 PM

@Dr. Furkan Gözükara when are you going to do anything that works with fp4/int4 for us peasants with 8GB VRAM lol

DDrocsid Imperator Maximus @Dr. Furkan Gözükara when are you going to do anything that works with fp4/int4 ...

Furkan Gözükara SECourses•12/15/24, 11:24 PM

8 gb vram already works

Furkan Gözükara SECourses•12/15/24, 11:24 PM

but you need 64 gb ram

FFurkan Gözükara SECourses 8 gb vram already works

Drocsid Imperator Maximus•12/15/24, 11:25 PM

Not with

SUPIR

SUPIR

lol

DDrocsid Imperator Maximus Not with `SUPIR` lol

Furkan Gözükara SECourses•12/15/24, 11:25 PM

FFurkan Gözükara SECourses but you need 64 gb ram

Drocsid Imperator Maximus•12/15/24, 11:25 PM

Oh, well... That sucks. I only have 16GB for now...

Drocsid Imperator Maximus•12/15/24, 11:25 PM

So, seems like I can't do nothing, eh?

DDrocsid Imperator Maximus So, seems like I can't do nothing, eh?

Furkan Gözükara SECourses•12/15/24, 11:25 PM

yep

Furkan Gözükara SECourses•12/15/24, 11:25 PM

best to upgrade 64 gb ram

FFurkan Gözükara SECourses yep

Drocsid Imperator Maximus•12/15/24, 11:25 PM

sad noises

Drocsid Imperator Maximus•12/15/24, 11:25 PM

Drocsid Imperator Maximus•12/15/24, 11:33 PM

Going to go the MassedCompute route temporarily then...

DDrocsid Imperator Maximus Going to go the MassedCompute route temporarily then...

Furkan Gözükara SECourses•12/15/24, 11:38 PM

nice

DDrocsid Imperator Maximus Going to go the MassedCompute route temporarily then...

Furkan Gözükara SECourses•12/15/24, 11:38 PM

instead of zip

Furkan Gözükara SECourses•12/15/24, 11:38 PM

downloader will download faster

Furkan Gözükara SECourses•12/15/24, 11:39 PM

use latest zip from patreon

FFurkan Gözükara SECourses nice

Drocsid Imperator Maximus•12/15/24, 11:39 PM

BTW, is there anyway I can pause an instance instead of having to delete it and start all over again?

FFurkan Gözükara SECourses use latest zip from patreon

Drocsid Imperator Maximus•12/15/24, 11:39 PM

Yeah, too late now lol. I thought the zip was easier anyhow, kinda.

DDrocsid Imperator Maximus BTW, is there anyway I can pause an instance instead of having to delete it and ...

Furkan Gözükara SECourses•12/15/24, 11:39 PM

good question. i think not . but lets ask @NicB

Drocsid Imperator Maximus•12/15/24, 11:41 PM

FWIW, your coupon no longer works (unless it was for A6000A6000 only, as I've now used an L40SL40S due to the available capacity) That said, I'm not complaining, the prices appear for now to be quite reasonable for what they are. I'm certain the unit economics may improve with more MAUs.

FFurkan Gözükara SECourses good question. i think not . but lets ask @NicB

Drocsid Imperator Maximus•12/16/24, 12:11 AM

I'm getting an error with

JoyCaption

JoyCaption

:
ImportError: tokenizers>=0.21,<0.22 is required for a normal functioning of this module, but found tokenizers==0.15.2.ImportError: tokenizers>=0.21,<0.22 is required for a normal functioning of this module, but found tokenizers==0.15.2.
Try: pip install transformers -U or pip install -e '.[dev]' if you're working with git mainTry: pip install transformers -U or pip install -e '.[dev]' if you're working with git main

DDrocsid Imperator Maximus FWIW, your coupon no longer works (unless it was for `A6000` only, as I've now u...

Furkan Gözükara SECourses•12/16/24, 12:11 AM

yes only for A6000 sadly

Furkan Gözükara SECourses•12/16/24, 12:11 AM

they gonna give new code for other gpus too

I am curious to maximise the PCs ability - with 6x 4090 and 512 DDR5 RAM (PCIe 5) - to train models

Similar Threads

Similar Threads

Similar Threads