Software Engineering Courses (SECourses)•14mo ago

cd kohya_ss git checkout d47e3e664286314c6b9728bb38b1c7a576211572

cd kohya_ss

git checkout d47e3e664286314c6b9728bb38b1c7a576211572

Rockerx•11/18/24, 12:42 PM

@Furkan Gözükara SECourses one question for understanding. why would 16GB config be slower than 12gb config. isn't more VRAM better? I am comparing configs and only difference is blocks to swap. should it now slow down training if 12gb config has more blocks to swap ?

FFurkan Gözükara SECourses you can request from kohya only he can do

Rockerx•11/18/24, 12:43 PM

so for v14 it is necessary to use the bat file fix ? or is it only if this error is encountered

Rockerx•11/18/24, 1:41 PM

also. 3rd question. do we still need Windows_update_kohya_and_fix_FLUX_Step2 ? or is it not necessary? @Furkan Gözükara SECourses

FFurkan Gözükara SECourses cd kohya_ss git checkout d47e3e664286314c6b9728bb38b1c7a576211572

Shamsi•11/18/24, 2:55 PM

I used these steps in the terminal at kohya_ss folder, still the same issue

RRockerx <@205854764540362752> one question for understanding. why would 16GB config be s...

Furkan Gözükara SECoursesOP•11/18/24, 4:24 PM

even kohya couldnt find it

Furkan Gözükara SECoursesOP•11/18/24, 4:24 PM

it is a weird thing

RRockerx so for v14 it is necessary to use the bat file fix ? or is it only if this error...

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

yes but bmaltais fixed

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

latest version should work now

RRockerx also. 3rd question. do we still need Windows_update_kohya_and_fix_FLUX_Step2 ? o...

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

not atm but in future maybe

SShamsi I used these steps in the terminal at kohya_ss folder, still the same issue

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

try latst version

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

maybe now works

Furkan Gözükara SECoursesOP•11/18/24, 4:25 PM

he made an update

FFurkan Gözükara SECourses try latst version

Shamsi•11/18/24, 4:56 PM

Sorry, what do you mean latest version? I'm V44 files from a day ago.

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭L40S, got it, thanks!

CMatias•11/18/24, 6:56 PM

L40S and A100 PCIe, are my first choices.
L40S is very polyvalent and has a good speed.
I use A100 PCIe when I don't want to loose time about searching for the parameters limits and fall into Out Of Memory problems.

SShamsi Sorry, what do you mean latest version? I'm V44 files from a day ago.

Furkan Gözükara SECoursesOP•11/18/24, 8:12 PM

i mean run updater bat file

Furkan Gözükara SECoursesOP•11/18/24, 8:12 PM

kohya gui fixed error

FFurkan Gözükara SECourses i mean run updater bat file

Shamsi•11/18/24, 8:18 PM

i'm running ubuntu on massed

SShamsi i'm running ubuntu on massed

Furkan Gözükara SECoursesOP•11/18/24, 8:20 PM

then run installer

Furkan Gözükara SECoursesOP•11/18/24, 8:20 PM

it will update

Furkan Gözükara SECoursesOP•11/18/24, 8:20 PM

our installer from zip file

FFurkan Gözükara SECourses our installer from zip file

Shamsi•11/18/24, 8:59 PM

link please doc

Furkan Gözükara SECoursesOP•11/18/24, 9:01 PM

https://www.patreon.com/posts/112099700

Furkan Gözükara SECoursesOP•11/18/24, 9:02 PM

use massed compute instructions

Shamsi•11/18/24, 9:23 PM

ty ty doc!

FFurkan Gözükara SECourses https://www.patreon.com/posts/112099700

Shamsi•11/18/24, 9:26 PM

I'm creating Loras

Shamsi•11/18/24, 9:26 PM

it is the same?

SShamsi I'm creating Loras

Furkan Gözükara SECoursesOP•11/18/24, 9:37 PM

yes

Furkan Gözükara SECoursesOP•11/18/24, 9:37 PM

uses same files

Shamsi•11/18/24, 9:38 PM

ok thank you again doc

Rockerx•11/18/24, 9:54 PM

thanks for the responses.

Rockerx•11/18/24, 9:55 PM

@Furkan Gözükara SECourses can we train FP8 models as well using same configs? or do I need to select fp8 in the training parameters. Atm I can train lora on FP8 using your fp8 config.

RRockerx <@205854764540362752> can we train FP8 models as well using same configs? or do...

Furkan Gözükara SECoursesOP•11/18/24, 10:01 PM

yes

Furkan Gözükara SECoursesOP•11/18/24, 10:01 PM

but there is no fp8 training for fine tuning

Furkan Gözükara SECoursesOP•11/18/24, 10:01 PM

kohya doesnt support

FFurkan Gözükara SECourses but there is no fp8 training for fine tuning

Rockerx•11/18/24, 11:29 PM

I was asking for the dreambooth tab. so fp8 is only for lora

RRockerx I was asking for the dreambooth tab. so fp8 is only for lora

Rockerx•11/18/24, 11:31 PM

@Furkan Gözükara SECourses do you know what is used for fp8 finetune ? do other tools like onetrainer support it ? or is the only option to do full training and then save as fp8

RRockerx <@205854764540362752> do you know what is used for fp8 finetune ? do other tools...

Furkan Gözükara SECoursesOP•11/18/24, 11:39 PM

it only converts base model into fp8

Furkan Gözükara SECoursesOP•11/18/24, 11:39 PM

training is still done as fp16

Furkan Gözükara SECoursesOP•11/18/24, 11:39 PM

kohya said me that he doesnt know how to make training in fp8

Shamsi•11/18/24, 11:53 PM

@Furkan Gözükara SECourses I did the updates as suggested and played with the repeat and epoch numbers. still it won't go past 1600 steps and 10 epochs (I've set it to 200). is this normal or am i missing something?

SShamsi <@205854764540362752> I did the updates as suggested and played with the repeat ...

Furkan Gözükara SECoursesOP•11/18/24, 11:57 PM

yes you need to enter max steps count

Furkan Gözükara SECoursesOP•11/18/24, 11:57 PM

in the gui look for it

FFurkan Gözükara SECourses yes you need to enter max steps count

Shamsi•11/19/24, 12:16 AM

Yes sirr

. Had to override the default =0.

SShamsi Yes sirr 🫡 . Had to override the default =0.

Furkan Gözükara SECoursesOP•11/19/24, 12:20 AM

yep

FFurkan Gözükara SECourses yes you need to enter max steps count

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/19/24, 6:54 AM

Perfect, just what I was gonna ask too

Got my answer quickly here. Thanks

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/19/24, 10:21 AM

Here's a challenge I want to complete. I have a fine-tuned checkpoint for my character, but I want my characters body to be taken from another character. So I trained a lora using pics of the other character, but the lora also includes that characters face. What do you guys suggest I do to get best result? I don't want my face changed because of this lora. So is using the <segment...> syntax in swarmui a way to go, or should I change the training images such that the head is cut off?

Appreciate this community so much, I learn something every day!

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭Here's a challenge I want to complete. I have a fine-tuned checkpoint for my cha...

Furkan Gözükara SECoursesOP•11/19/24, 10:45 AM

i dont think it will work

Furkan Gözükara SECoursesOP•11/19/24, 10:45 AM

you can inpaint face of another character

Rockerx•11/19/24, 11:46 AM

@Furkan Gözükara SECourses reporting 2 things.

now when I install v44 or v14 scripts it does update kohya and I am getting correct step count.
in your lora configs you are using additional parameters for block swap. this throws an exit 0 error and training exits. the UI in lora in latest kohya now has block swap option in UI. if I manually set it to what your config has and remove the additional parameter the training works but not sure if as the UI option says (use with fused back pass) which is not available in UI. but the training works with block swap if I set in UI and remove your addition parameters setting.
Can you fix the configs so that it works for lora ? have not tested the finetune in new config but see if that also needs update etc.

wardensc2•11/20/24, 1:57 AM

hi @Furkan Gözükara SECourses I just download dreambooth v15 few days ago now you already uploaded v16. Is there something new in this new version ?

cd kohya_ss git checkout d47e3e664286314c6b9728bb38b1c7a576211572

Similar Threads

Similar Threads

Similar Threads