Software Engineering Courses (SECourses)•16mo ago

@Dr. Furkan Gözükara I'm still getting OOM errors when trying to train a lora on 12GB, i'm all the

@Dr. Furkan Gözükara I'm still getting OOM errors when trying to train a lora on 12GB, i'm all the way down to rank 8, i have 11.2 GB of Vram available and still getting it

TheToday•9/9/24, 6:55 PM

Anyone else having trouble with Forge? I'm using the same old thing in Flux and I get OutOfMemory error in Massedcompute with the 48gb

TheToday•9/9/24, 7:06 PM

I think it's when patching Lora, but it worked perfectly a few days ago

Zono50OP•9/9/24, 7:38 PM

can we use our lora with gguf quantized models of dev?

TheToday•9/9/24, 8:21 PM

the problem is with --always-gpu @Dr. Furkan Gözükara

Zono50OP•9/9/24, 8:22 PM

what's the recommended way to run inference with low gpu cards? we can't use flux1.dev

Kallamamran•9/9/24, 8:33 PM

Trained a lora with AI-toolkit and the outputs look good, but when I try to run it in Forge it doesn't at all look like the results in Ia-toolkit :/

Kallamamran•9/9/24, 8:33 PM

What am I missing

Kallamamran•9/9/24, 8:34 PM

I have trained on fp16 dev and are usiing dev nf4 in forge. Is that why?

Kallamamran•9/9/24, 8:34 PM

Can't trained LORA's be run in Forge? -

ZZono50 oh ok gotcha, i have 12 GB of vram and 64 GB of system ram, i am watching your v...

Furkan Gözükara SECourses•9/9/24, 8:43 PM

awesome

ZZono50 @Dr. Furkan Gözükara I'm still getting OOM errors when trying to train a lora o...

Furkan Gözükara SECourses•9/9/24, 8:43 PM

which config you using?

ZZono50 can we use our lora with gguf quantized models of dev?

Furkan Gözükara SECourses•9/9/24, 8:44 PM

i havent tried yet

Furkan Gözükara SECourses•9/9/24, 8:44 PM

but swarm ui should work out of box with fp8

KKallamamran I have trained on fp16 dev and are usiing dev nf4 in forge. Is that why?

Furkan Gözükara SECourses•9/9/24, 8:44 PM

nf4 is the problem

ZZono50 what's the recommended way to run inference with low gpu cards? we can't use flu...

Furkan Gözükara SECourses•9/9/24, 8:44 PM

swarmui

Furkan Gözükara SECourses•9/9/24, 8:44 PM

most optimized

rafraf•9/9/24, 8:46 PM

hey @Dr. Furkan Gözükara using swarmui grid generator, if run a grid gen with two params, and you rerun the same params but add an extra value, does it overwrite the entire previous gen? I was under the impression that simply adding a value in an existing param, just inserts the extra gens into the existing grid

Rrafraf hey @Dr. Furkan Gözükara using swarmui grid generator, if run a grid gen with tw...

Furkan Gözükara SECourses•9/9/24, 8:46 PM

i am not sure

Furkan Gözükara SECourses•9/9/24, 8:46 PM

i think it has mixed behaviour

rafraf•9/9/24, 8:47 PM

its really comfusing. using on runpod. ive tested on smalll images to see the mechanism, and it seems sometimes erratic, unpredictable. i am almost sure this the case

rafraf•9/9/24, 8:59 PM

but one thing i d like to know is, if i interrupt the grid genartion, can i re-run it? wi ll it skip already generated images? Or does the interruptioon corrupt the files? Surely you've tested and experienced the behavior?

Rrafraf but one thing i d like to know is, if i interrupt the grid genartion, can i re-r...

Furkan Gözükara SECourses•9/9/24, 9:00 PM

YES

Furkan Gözükara SECourses•9/9/24, 9:00 PM

it can resume

Zono50OP•9/9/24, 9:00 PM

@Dr. Furkan Gözükara in kohya can we puase and resume traininngs?

ZZono50 @Dr. Furkan Gözükara in kohya can we puase and resume traininngs?

Furkan Gözükara SECourses•9/9/24, 9:03 PM

currently it doesnt work i think

Furkan Gözükara SECourses•9/9/24, 9:03 PM

with fine tuning

Furkan Gözükara SECourses•9/9/24, 9:03 PM

you can use last check point but people said

FFurkan Gözükara SECourses nf4 is the problem

Kallamamran•9/9/24, 9:03 PM

Noticed

Furkan Gözükara SECourses•9/9/24, 9:03 PM

lora doesnt work

Furkan Gözükara SECourses•9/9/24, 9:03 PM

i didnt test

Zono50OP•9/9/24, 9:03 PM

ah ok

Furkan Gözükara SECourses•9/9/24, 10:26 PM

time to make some image show :d

Furkan Gözükara SECourses•9/9/24, 10:59 PM

40 images posted here : https://medium.com/@furkangozukara/ultimate-flux-lora-training-tutorial-windows-and-cloud-deployment-abb72f21cbf8

Medium

Ultimate FLUX LoRA Training Tutorial: Windows and Cloud Deployment

I have done total 104 different LoRA trainings and compared each one of them to find the very best hyper parameters and the workflow for…

Phuong Bui•9/9/24, 11:10 PM

How to emphasize keywords that need training in the prompt? Would it be more emphasized if I caplock the keyword or put it in ""?

PPhuong Bui How to emphasize keywords that need training in the prompt? Would it be more emp...

Furkan Gözükara SECourses•9/9/24, 11:11 PM

you mean improve their importance?

Furkan Gözükara SECourses•9/9/24, 11:11 PM

i think you can write multiple times

Furkan Gözükara SECourses•9/9/24, 11:11 PM

but it depends on model

Furkan Gözükara SECourses•9/9/24, 11:11 PM

for flux it has internal captioning works amazing

FFurkan Gözükara SECourses i think you can write multiple times

Phuong Bui•9/9/24, 11:14 PM

That's right, I want to increase the importance of keywords. I'm training on flux to see if capping or leaving it in "" is effective. Or just need each image to have a keyword?

PPhuong Bui That's right, I want to increase the importance of keywords. I'm training on flu...

Furkan Gözükara SECourses•9/9/24, 11:18 PM

good question sadly i didnt test

Furkan Gözükara SECourses•9/9/24, 11:18 PM

flux doesnt have a research paper

Furkan Gözükara SECourses•9/9/24, 11:18 PM

so only way is testing

Phuong Bui•9/9/24, 11:25 PM

I understand, let me try testing with each image having the same keyword

МИШАНЯ•9/10/24, 12:50 AM

Is there a flux IP ADAPTER FACE?
I want to transfer an existing (real face) Into a cartoon style...

ММИШАНЯ Is there a flux IP ADAPTER FACE? I want to transfer an existing (real face) Into...

Furkan Gözükara SECourses•9/10/24, 12:52 AM

i heard someone trying but didnt saw yet

FFurkan Gözükara SECourses i heard someone trying but didnt saw yet

МИШАНЯ•9/10/24, 12:53 AM

If there is information, where can I find it? I really need it.((

ММИШАНЯ If there is information, where can I find it? I really need it.((

Furkan Gözükara SECourses•9/10/24, 12:54 AM

i dont know atm

FFurkan Gözükara SECourses i dont know atm

sergio_gonzo•9/10/24, 1:11 AM

dr furkan I have a question I used the rank 5 setting on a first model with 200 epoch and the result was stellar I tried another model with the same amount of images, another character and the result was good but not as good as the one before this one I only put 125 epoch can that change in value affect the final result? I ask this because the first one, the one that was real good, the best lora was 75 I thought that if I lowered the epoch it would result in having a good checkpoint before epoch 100 is that sum wrong? is it better to leave the 200 epoch as the rank 5 setting commands?

AIGambino•9/10/24, 1:11 AM

What are rank meanings in flux training?

Ssergio_gonzo dr furkan I have a question I used the rank 5 setting on a first model with 200 ...

Furkan Gözükara SECourses•9/10/24, 1:11 AM

i think best is go up to 200 and compare check points

@Dr. Furkan Gözükara I'm still getting OOM errors when trying to train a lora on 12GB, i'm all the

Similar Threads