Software Engineering Courses (SECourses)•13mo ago

SwarmUi\models\Diffusion Models\a

Furkan Gözükara SECoursesOP•12/17/24, 1:48 AM

SwarmUi\models\Diffusion Models\b

Furkan Gözükara SECoursesOP•12/17/24, 1:48 AM

SwarmUi\models\Diffusion Models\c

QM•12/17/24, 1:57 AM

I was running two trainings on two individuals, the saved epochs are blending faces of the two people - because my models were saved in the same output folder

Zzaiar hey quick question, im trying to train a LoRa on Flux, I have a 3060 with 12gb s...

zaiar•12/17/24, 2:17 AM

i have 32gb of RAM as well if that clarifies, any help would be greatly appreciated

very new to this

FFurkan Gözükara SECourses SwarmUi\models\Diffusion Models\a

QM•12/17/24, 2:18 AM

I saved every 14 Epoch - the first epoch was ready so i tested it - but it is not generating anything near the original people

QM•12/17/24, 2:42 AM

The second model after EPOC 28 has gone even further astray

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•12/17/24, 10:41 AM

best caption type for sdxl? training prompts in your joy caption app?

Zzaiar i have 32gb of RAM as well if that clarifies, any help would be greatly apprecia...

Furkan Gözükara SECoursesOP•12/17/24, 11:31 AM

yes please upgrade to 64 gb and all training errors will be fixed

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭best caption type for sdxl? training prompts in your joy caption app?

Furkan Gözükara SECoursesOP•12/17/24, 11:31 AM

pre alpha good

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•12/17/24, 11:32 AM

hm? it's either descriptive, rnd tags or training-prompts.

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭hm? it's either descriptive, rnd tags or training-prompts.

Furkan Gözükara SECoursesOP•12/17/24, 11:33 AM

i would say it depends on your base model

Furkan Gözükara SECoursesOP•12/17/24, 11:34 AM

make few comparison what model generates for each type

FFurkan Gözükara SECourses yes please upgrade to 64 gb and all training errors will be fixed

zaiar•12/17/24, 1:56 PM

its a little bit out of my budget right now, I just bought the 3060 to upgrade from 8 to 12gb of VRAM, and then I bought 2 more 8gb of RAM to increase from 16gb to 32. is it possible to complete training with 32gb of ram? even if it takes a long time

zaiar•12/17/24, 1:59 PM

I monitor my memory usage during training, my RAM stays at about 25-28GB of usage through-out the process, but seems like my GPU is maxing out, it'll hit 15.1gb of VRAM usage and then 2 seconds later the training crashes

zaiar•12/17/24, 1:59 PM

ill run the training and copy paste the cmd message later, not home right now

Timson•12/17/24, 3:01 PM

@zaiar just do your AI research in the cloud, it's not worth upgrading local PC to match specific requirements unless you are planing to do it as a full time occupation or you have other good uses for it (like gaming PC).
6.5$ gives you a whole work day access to a monster configuration server. It's good if you just want to experiment from time to time and requires no long term investments

Timson•12/17/24, 3:06 PM

@Furkan Gözükara SECourses what is the maximum number of dreambooth training steps you performed on flux-dev until you encountered serious model sanity degradation? I'm starting to see 6 finger hands on 3000+ steps. Can regularisation help with fine tuning model to a more complex concepts (like entire movie franchise visual style, instead of specific characters)?

Yobin•12/17/24, 5:16 PM

@Furkan Gözükara SECourses what's your opinion on doing dreambooth flux training with style then use the checkpoint dreambooth fine tuned for character lora training? I want to seperate the style and character models

Jassim•12/17/24, 5:35 PM

@Furkan Gözükara SECourses im trying to use kohya ss to finetune sdxl on runpod using your method its taking forever for step1/2 and i dont even know if its even still running or stopped

Testitou•12/17/24, 8:14 PM

On my side, after extracting to LORA with rank 640 and 512, and converting to fp8, swarmui render black image. With rank 128 / 256 and 384,fp8 it's OK.

TTimson <@205854764540362752> what is the maximum number of dreambooth training steps yo...

Furkan Gözükara SECoursesOP•12/17/24, 8:53 PM

i did over 50000 samples training

Furkan Gözükara SECoursesOP•12/17/24, 8:53 PM

https://huggingface.co/MonsterMMORPG/Big_FLUX_Fine_Tuning_Experiments/tree/main

MonsterMMORPG/Big_FLUX_Fine_Tuning_Experiments at main

JJassim <@205854764540362752> im trying to use kohya ss to finetune sdxl on runpod using...

Furkan Gözükara SECoursesOP•12/17/24, 8:54 PM

runpod initial load and setup sucks so bad

TTestitou On my side, after extracting to LORA with rank 640 and 512, and converting to fp...

Furkan Gözükara SECoursesOP•12/17/24, 8:54 PM

interesting

Renzo•12/18/24, 2:17 AM

What settings are available to increase the speed of Dreambooth training? I'm very interested in experimenting with different configurations even if quality takes a hit. My understanding is that the configurations included in Kohya_FLUX_DreamBooth_v17 are meant to provide ways to fine tune models on different types of GPUs (vRAM).

RRenzo What settings are available to increase the speed of Dreambooth training? I'm ve...

Furkan Gözükara SECoursesOP•12/18/24, 8:27 AM

yes

Furkan Gözükara SECoursesOP•12/18/24, 8:28 AM

if you want speed only way is reducing resolution for fine tuning

Furkan Gözükara SECoursesOP•12/18/24, 8:28 AM

which will reduce your quality

Renzo•12/18/24, 11:53 AM

Right. Or reducing the number of training photos.

Renzo•12/18/24, 12:38 PM

@Furkan Gözükara SECourses What about optimizers and other variables that could make the training converge to a good result quicker? Is Adafactor the best optimizer based on your testing?

QM•12/18/24, 8:17 PM

is there a particular reason behind using ohwx man?

QM•12/18/24, 8:17 PM

or are these just any random characters - with the objective to not use any characters that might have been present

RRenzo <@205854764540362752> What about optimizers and other variables that could make ...

Furkan Gözükara SECoursesOP•12/19/24, 12:54 AM

Adafactor best for memory

Furkan Gözükara SECoursesOP•12/19/24, 12:54 AM

each one requires a new research

QQM is there a particular reason behind using ohwx man?

Furkan Gözükara SECoursesOP•12/19/24, 12:54 AM

ohwx is a rare token for clip

Furkan Gözükara SECoursesOP•12/19/24, 12:54 AM

so i still use it

Hiramos_•12/19/24, 4:13 AM

@Furkan Gözükara SECourses Is it possible to perform DreamBooth training on a custom Flux checkpoint?

HHiramos_<@205854764540362752> Is it possible to perform DreamBooth training on a custom ...

Furkan Gözükara SECoursesOP•12/19/24, 9:44 AM

yes

Furkan Gözükara SECoursesOP•12/19/24, 9:44 AM

https://medium.com/@furkangozukara/thoroughly-experimented-with-fine-tuning-dreambooth-training-of-flux-dev-de-distill-pixelwave-13fab8c625bf

Medium

Thoroughly experimented with Fine-Tuning / DreamBooth training of F...

Thoroughly experimented with Fine-Tuning / DreamBooth training of Flux-dev-de-distill, PixelWave v03, Verus Vision and base FLUX Dev model…

Zhigarev•12/19/24, 11:38 AM

Does anyone here have experience creating the same DreamBooth character with different outfits? The problem is that when training in DreamBooth without captions, it's necessary to specify every clothing element to achieve the desired look, which heavily consumes tokens in the prompt and reduces adherence to the prompt during generation. Do you have any experience with this?

D4UNTLESS•12/19/24, 12:09 PM

@Furkan Gözükara SECourses hi
I'm about to do flux fine tuning on massed compute by following your tutorial but I have two questions:

1. What's the difference between a model fine tuned by using batch size of 1 and a model fine tuned by using a batch size of 7? How much (by percentage) is the 1 batch size better than the 7 batch size?
2. I'm fine tuning the model to train a realistic character. I have 410 training images. How many epochs should I go with if I use a batch size of 7? I need your best guess. In your tutorial, you mentioned that for 256 training images, 40 epoch is best for a batch size of 7

DD4UNTLESS <@205854764540362752> hi I'm about to do flux fine tuning on massed compute by f...

Furkan Gözükara SECoursesOP•12/19/24, 12:45 PM

very small differnce

Furkan Gözükara SECoursesOP•12/19/24, 12:45 PM

go with up to 50 epochs

Furkan Gözükara SECoursesOP•12/19/24, 12:45 PM

and compare

FFurkan Gözükara SECourses Adafactor best for memory

Renzo•12/19/24, 12:52 PM

Interesting! I will experiment with AdamW today.

DD4UNTLESS <@205854764540362752> hi I'm about to do flux fine tuning on massed compute by f...

Timson•12/19/24, 2:33 PM

in my experience the more images are in the training set the more training steps is required to reach the same level of training.
For example 15 character image set is already good at 1800 steps (120 epochs), but with 50 images in training set the same level of training is achieved around 3200 steps (64 epochs)

Timson•12/19/24, 2:38 PM

SO the more images there are - the more generalised concept the model is learning - the more training steps required

Timson•12/19/24, 2:42 PM

IMO the sweet spot for character training is 25 images. With more images the training gets exponentially slower and the end result is not necessarily better

TTimson in my experience the more images are in the training set the more training steps...

Furkan Gözükara SECoursesOP•12/19/24, 3:35 PM

true

TTimson IMO the sweet spot for character training is 25 images. With more images the tra...

Furkan Gözükara SECoursesOP•12/19/24, 3:35 PM

it depends your aims

SwarmUi\models\Diffusion Models\a

Similar Threads

Similar Threads

Similar Threads