Software Engineering Courses (SECourses)•15mo ago

@Dr. Furkan Gözükara on flux-dev-de-distill to get resemblance with regularization I used a much hi

@Dr. Furkan Gözükara on flux-dev-de-distill to get resemblance with regularization I used a much higher lr 0.004 on 160 epochs looks still a little undertrained, my lora rank and alpha is set like this, this can be a problem

Ddsienra @Dr. Furkan Gözükara on flux-dev-de-distill to get resemblance with regularizati...

Furkan Gözükara SECourses•10/12/24, 10:55 PM

yep

Furkan Gözükara SECourses•10/12/24, 10:55 PM

you basically changed the LR i researched

Furkan Gözükara SECourses•10/12/24, 10:55 PM

reduced it like 1/8

Furkan Gözükara SECourses•10/12/24, 10:55 PM

i am offline

FFurkan Gözükara SECourses yep

dsienraOP•10/12/24, 10:57 PM

ohhhh, now I understands

dsienraOP•10/12/24, 11:01 PM

No wonder it was taking me so long to train my LoRAs, the learning rate was too low

dsienraOP•10/12/24, 11:16 PM

@Dr. Furkan Gözükara sorry that you are offline, no hurry to answer, in your best configs nowhere says if the config is a finetune or a lora

Ddsienra @Dr. Furkan Gözükara sorry that you are offline, no hurry to answer, in your bes...

Error-404-unknown•10/13/24, 12:08 AM

LoRas are in Best_config folder inside kohya_gui_flux_installer.zip and finetunes are in Best_config_v5.zip

Ddsienra the loras I'm training on flux-dev-de-distill works perfect on regular flux-dev,...

hamuz•10/13/24, 12:12 AM

It's great to hear that! I just started training a LoRA about an hour ago, but I canceled it now because I'm going to rent 2 A100s for the training. Do you have any parameters that you recommend adjusting between De-Distilled and Standard flux LoRA training?

Hhamuz It's great to hear that! I just started training a LoRA about an hour ago, but I...

dsienraOP•10/13/24, 12:25 AM

you can use the same settings, the samples during training doesn't work, other than that the same, you can train multiple subjects in one lora with the de-distilled model and learn new concepts much better

EError-404-unknown LoRas are in Best_config folder inside kohya_gui_flux_installer.zip and finetune...

dsienraOP•10/13/24, 12:26 AM

thanks!!

Ddsienra you can use the same settings, the samples during training doesn't work, other t...

hamuz•10/13/24, 12:42 AM

Thank you so much

Ddsienra The training was finished, resemblance is almost perfect, but it still looks a l...

druhl•10/13/24, 1:32 AM

Can you resume/ increase the no. of epochs in the existing training? Or is it that you had paused it?

Ddsienra @Dr. Furkan Gözükara on flux-dev-de-distill to get resemblance with regularizati...

druhl•10/13/24, 1:41 AM

Isn't 0.004 faster than 0.000004? Or is LR counter opposite in ML?

druhl•10/13/24, 4:22 AM

Sadly, my pose training has completely failed on both the models. Think I'm doing something wrong on the fundamental level itself

, 'cuz I'm approaching it like a character fine-tune.

Ddsienra Nice, If you have to train more than one subject it is your only choice, in my t...

druhl•10/13/24, 5:28 AM

Get what you're trying to say. Changing strategy, training different poses as multiple separate concepts instead of training them under the same poses class.

druhl•10/13/24, 5:29 AM

Inference on dev_distill can be pretty good as well, if you tweak it in Comfy. But yes, it is slower.

druhl•10/13/24, 5:31 AM

flux_dev versus flux_dev_de_distill

druhl•10/13/24, 5:32 AM

versus Pro 1.1

Ddruhl Can you resume/ increase the no. of epochs in the existing training? Or is it th...

dsienraOP•10/13/24, 5:51 AM

you can get any lora checkpoint and resume the training from there, there is no pause option.

Ddruhl Isn't 0.004 faster than 0.000004? Or is LR counter opposite in ML?

dsienraOP•10/13/24, 6:02 AM

0.0004 is much faster but it was my mistake, my lora rank was at 128 and my network alpha at 16 that means that I had to increase the lr up to 0.0004 to match the dr. Furkan settings, just use one of his lora configs and you are good to go. lora rank 128 alpha 1 or rank 128 alpha 128 is the same setting, because 128 divided 128 equals 1, all this iinfo is kohya's documentation

Ddruhl Get what you're trying to say. Changing strategy, training different poses as mu...

dsienraOP•10/13/24, 6:19 AM

unless the poses you are training are very uncommon may be the pose is already trained on the base model, you have to prompt it correctly, if the pose is kind of unique then you can have a dataset of pictures with your especial pose and train it with captions like "nameofthepose pose"

Ddsienra unless the poses you are training are very uncommon may be the pose is already t...

druhl•10/13/24, 6:22 AM

That's how I'm captioning my images right now to start new training. But what about unique token and class token? Do I need to enter all of them comma seperated or I can just go into the kohya directory directly & modify folder names like 1_nameofthefirstpose, 1_nameofthesecondpose, and so on ?

FFurkan Gözükara SECourses yes normal

deepblue_1337•10/13/24, 6:28 AM

Thank you! Would a 4060 be much faster?

Any other way to speed up?

Ddruhl That's how I'm captioning my images right now to start new training. But what ab...

dsienraOP•10/13/24, 6:29 AM

you can just change the folder name, you can use an unique token an space pose, so is in the class poses that the model already knows for example "pose1 pose" or name of the folder "1_pose1 pose" just a unique token without the class pose can confuse the model, that's my theory, I never trained a pose...

Ddsienra you can just change the folder name, you can use an unique token an space pose, ...

druhl•10/13/24, 6:34 AM

All great info, thanks a lot!

Ddruhl All great info, thanks a lot! 🙂

dsienraOP•10/13/24, 6:44 AM

you welcome!!

m3dia_offline•10/13/24, 7:07 AM

I am having a lot of problems trying to run my offline comfyui workflow on runpod.

Missing custom nodes via manager are not installing even after pod resets and manually resetting the pod via jupiter lab, can anyone on here give me some assistance in debugging this? I can send you a video recording of the whole thing for ease of understanding

Ddsienra @Dr. Furkan Gözükara sorry that you are offline, no hurry to answer, in your bes...

Furkan Gözükara SECourses•10/13/24, 10:08 AM

currently fine tunings are best

Furkan Gözükara SECourses•10/13/24, 10:08 AM

i have completed testing

Furkan Gözükara SECourses•10/13/24, 10:08 AM

going to update patreon posts today

Ddruhl Isn't 0.004 faster than 0.000004? Or is LR counter opposite in ML?

Furkan Gözükara SECourses•10/13/24, 10:09 AM

0.004 is 1000 times bigger LR

Mm3dia_offline I am having a lot of problems trying to run my offline comfyui workflow on runpo...

Furkan Gözükara SECourses•10/13/24, 10:11 AM

i can try to help if you upgrade to gold membership

Furkan Gözükara SECourses•10/13/24, 10:27 AM

best checkpoints of 8 different test

Furkan Gözükara SECourses•10/13/24, 10:28 AM

bs7 and bs1 getting same epoch as best means both has got perfect LR

Furkan Gözükara SECourses•10/13/24, 10:31 AM

some big testing time :

deepblue_1337•10/13/24, 10:41 AM

if my computer unexpectedly reboots while training, is there a possibillity to resume process or do i jave to start over?

Ddeepblue_1337 if my computer unexpectedly reboots while training, is there a possibillity to r...

Furkan Gözükara SECourses•10/13/24, 10:41 AM

you have to restart

Furkan Gözükara SECourses•10/13/24, 10:41 AM

you can use latest fine tuning checkpoint

Furkan Gözükara SECourses•10/13/24, 10:41 AM

for lora there is lora weights but i dont know if working

Furkan Gözükara SECourses•10/13/24, 10:42 AM

if it is running on massed cokmpute or runpod, you can even close your computer

Furkan Gözükara SECourses•10/13/24, 10:42 AM

they run on cloud

deepblue_1337•10/13/24, 10:48 AM

you mention somewhere that you have a discount on massed compute, is that right? if yes how/where?

thanks

Ddeepblue_1337 you mention somewhere that you have a discount on massed compute, is that right?...

Furkan Gözükara SECourses•10/13/24, 10:49 AM

select A6000 GPU

Furkan Gözükara SECourses•10/13/24, 10:49 AM

use SECourses coupon

deepblue_1337•10/13/24, 10:56 AM

did you ever use fluxgym? As far as I understand it does also use Kohya, but has a (more or less) one-klick setup?

deepblue_1337•10/13/24, 10:58 AM

or any thoughts on providing a script that installs all dependencies and starts the program by yourself? I am a computer engineer (by training) myself, though out of the industry for >20years, and the setup usage could be made massively easier (especially for non-technical people)

deepblue_1337•10/13/24, 10:58 AM

just as suggestions, otherwise your content is great

FFurkan Gözükara SECourses 0.004 is 1000 times bigger LR

druhl•10/13/24, 11:48 AM

I associated bigger with faster. He said his checkpoint was taking too much time to train, but I assumed if I use 0.0004 instead of 0.000004, I train faster.

druhl•10/13/24, 11:49 AM

But then he mentioned it also has something to do with his lora rank selection of 128.

@Dr. Furkan Gözükara on flux-dev-de-distill to get resemblance with regularization I used a much hi

Similar Threads

Similar Threads

Similar Threads