Software Engineering Courses (SECourses)•15mo ago

using embeddings or TE training might solve that issue though. Not bad on the left with embedding, o

using embeddings or TE training might solve that issue though. Not bad on the left with embedding, on the right just "woman" with no bleeding.

Ddxqb using embeddings or TE training might solve that issue though. Not bad on the le...

Furkan Gözükara SECourses•10/6/24, 8:18 PM

TE training doesnt solve

EyeSpyBekoAI•10/6/24, 8:18 PM

hey @Dr. Furkan Gözükara quick question. Can Flux 1.1 be Lora trained or even downloaded? As far as I see it can only be used in platforms at the moment.

Furkan Gözükara SECourses•10/6/24, 8:18 PM

actually i trained single layer

Furkan Gözükara SECourses•10/6/24, 8:18 PM

it didnt solve too :d

EEyeSpyBekoAI hey @Dr. Furkan Gözükara quick question. Can Flux 1.1 be Lora trained or even do...

Furkan Gözükara SECourses•10/6/24, 8:19 PM

nope they didnt publish any model

FFurkan Gözükara SECourses TE training doesnt solve

dxqbOP•10/6/24, 8:19 PM

no it doesn't. it might solve MY problem that the "instance" is harder to train using regularization. it doens't solve yours. I don't know what's going on with yours. wanna share config+pics?

Ddxqb no it doesn't. it might solve MY problem that the "instance" is harder to train ...

Furkan Gözükara SECourses•10/6/24, 8:19 PM

yes my config same as on patreon nothing different

same as kohya one

Furkan Gözükara SECourses•10/6/24, 8:20 PM

my dataset here

FFurkan Gözükara SECourses nope they didnt publish any model

EyeSpyBekoAI•10/6/24, 8:20 PM

I didnt' think so either, but then I just came across this? Looks like pro 1.1 is available?
https://civitai.com/models/618692?modelVersionId=922358

FFurkan Gözükara SECourses yes my config same as on patreon nothing different 😄 same as kohya one

dxqbOP•10/6/24, 8:20 PM

since we are looking for a specific problem I'd prefer to have original data. config file + image files. DM

EEyeSpyBekoAI I didnt' think so either, but then I just came across this? Looks like pro 1.1 i...

Furkan Gözükara SECourses•10/6/24, 8:21 PM

i think only inference via api

Ddxqb since we are looking for a specific problem I'd prefer to have original data. c...

Furkan Gözükara SECourses•10/6/24, 8:21 PM

sure sending dm

FFurkan Gözükara SECourses both fast and good on 24 gb gpus

B0nd•10/6/24, 8:23 PM

whats the minium steps needed you think for training 40 images, for true photorealism?

BB0nd whats the minium steps needed you think for training 40 images, for true photore...

Furkan Gözükara SECourses•10/6/24, 8:23 PM

lets say 100 epoch for good resemblance

Furkan Gözükara SECourses•10/6/24, 8:23 PM

4000 steps batch size 1

FFurkan Gözükara SECourses 4000 steps batch size 1

B0nd•10/6/24, 8:24 PM

great thanks

Furkan Gözükara SECourses•10/6/24, 8:25 PM

i will update batch size 1 fine tuning configs

Furkan Gözükara SECourses•10/6/24, 8:25 PM

for latest kohya

B0nd•10/6/24, 8:31 PM

when training lora did you notice much difference between adafactor and adamw8bit?

BB0nd when training lora did you notice much difference between adafactor and adamw8bi...

Furkan Gözükara SECourses•10/6/24, 8:32 PM

i plan to investigate

Furkan Gözükara SECourses•10/6/24, 8:32 PM

newest optimizers hopefully soon

Furkan Gözükara SECourses•10/6/24, 8:32 PM

but each one would require different parameters

Furkan Gözükara SECourses•10/6/24, 8:45 PM

hahaha 4765 MB can do flux fine tuning now :d

FFurkan Gözükara SECourses hahaha 4765 MB can do flux fine tuning now :d

crystalwizard•10/6/24, 8:46 PM

except it cant. you know what dpo is?

Ccrystalwizard except it cant. you know what dpo is?

Furkan Gözükara SECourses•10/6/24, 8:47 PM

it is i am doing right now. for single concept it will work perfect

crystalwizard•10/6/24, 8:47 PM

you can't fine tune, or train, flux. you are just creating something that gives you an image but you're not updating the weights or training the model, Doctor

Ccrystalwizard you can't fine tune, or train, flux. you are just creating something that gives ...

Furkan Gözükara SECourses•10/6/24, 8:49 PM

if i am not updating weights of the model it must be magic then

FFurkan Gözükara SECourses if i am not updating weights of the model it must be magic then 😄

crystalwizard•10/6/24, 8:49 PM

i'll ask you again, do you know what DPO is

Ccrystalwizard i'll ask you again, do you know what DPO is

Furkan Gözükara SECourses•10/6/24, 8:50 PM

tell me full words of it

MMaxivi 128x128 and 3e-06

dxqbOP•10/6/24, 8:50 PM

did you try man or woman

FFurkan Gözükara SECourses tell me full words of it

crystalwizard•10/6/24, 8:50 PM

the paper's on arXIv - i suggest you read it

Ccrystalwizard you can't fine tune, or train, flux. you are just creating something that gives ...

dxqbOP•10/6/24, 9:01 PM

that's hilarious statement.

dxqbOP•10/6/24, 9:04 PM

if you know so much about how Flux dev was created, please tell us how exactly the distillation from pro to dev was done, and how DPO was applied (which I'm not doubting - why wouldn't they)

FFurkan Gözükara SECourses ram usage shouldnt be much unless your vram is low

Ereshkigal•10/6/24, 9:11 PM

OK. Just asking cuz I have a 4090 & I've seen upwards of 60GB of system RAM usage when not rendering, when I start a render it goes down (I'm guessing that's loaded on to the gpu).

Ddxqb if you know so much about how Flux dev was created, please tell us how exactly t...

crystalwizard•10/6/24, 9:11 PM

you don't apply dpo to a base model - not if you're smart

dxqbOP•10/6/24, 9:13 PM

tell us why. DPO is proposed as an alternative to RLHF which obviously was applied very successfully for LLMs

Furkan Gözükara SECourses•10/6/24, 9:19 PM

fine tuning configs for 6 gb, 8gb, 10 gb , 12 gb , 16gb and 24 gb

Furkan Gözükara SECourses•10/6/24, 9:19 PM

Ddxqb tell us why. DPO is proposed as an alternative to RLHF which obviously was appl...

crystalwizard•10/6/24, 9:21 PM

I'll let you go read the paper and think

dxqbOP•10/6/24, 9:22 PM

so you don't know the answer. alright.

Ddxqb so you don't know the answer. alright.

crystalwizard•10/6/24, 9:37 PM

i do, but i don't feel like hand-feeding you. and you trying to manipulate me into handing you the research you don't want to do isn't going to work

dxqbOP•10/6/24, 9:38 PM

you are making unsubstantiated claims that you know more about how Flux was created than the general public. anyone can read papers. you have to substantiate your claim.

Ddxqb you are making unsubstantiated claims that you know more about how Flux was crea...

crystalwizard•10/6/24, 9:40 PM

i owe you nothing. i direct you, if you want a place to start your research other than arXiv, to all the posts on this discord ALONE where people talk about how many times they've tried to train flux for something and how it doesn't work. even Furkan himself has expressed issues with that. and that is all i'm giving you for free. go do the research

crystalwizard•10/6/24, 9:40 PM

be just as snarky as you like now, i'm done with you

dxqbOP•10/6/24, 9:41 PM

blocked

crystalwizard•10/6/24, 9:44 PM

blocking on discord is funny. you still see that the other person has posted something, but you can't see what. you can't tell if they're talking about you or not. blocking on discord is a good way to drive yourself crazy. and when you're the one that's tried to start an arguement, and then block the person you were trying to argue with because they wouldn't play your game, that says something about you to everyone watching

B0nd•10/6/24, 9:45 PM

Whats the best upscaler model that allows commercial use?

BB0nd Whats the best upscaler model that allows commercial use?

crystalwizard•10/6/24, 9:45 PM

topaz's models

Ccrystalwizard topaz's models

B0nd•10/6/24, 9:46 PM

can they be intergrated into comfyUI?

BB0nd can they be intergrated into comfyUI?

crystalwizard•10/6/24, 9:47 PM

no. they're propiratory and exclusive to topaz, and not free - but they are the best and you can use the output commercially

using embeddings or TE training might solve that issue though. Not bad on the left with embedding, o

Similar Threads