Software Engineering Courses (SECourses)•17mo ago

My first lora using @Dr. Furkan Gözükara tutorial and config came out great. Using Forge it is work

My first lora using @Dr. Furkan Gözükara tutorial and config came out great.

Using Forge it is working with the full flux dev model but not the fp8. Is this a forge thing?

FFurkan Gözükara SECourses if you need speed Rank_2_27360MB_Fast.json and batch size 2

GomezBro•8/31/24, 4:51 AM

Couldnt I technically do like RANK 6 Batch 4 or 5 on 1 A6000? - Currently doing rank 2 FAST Batch 2 on 1 A6000, taking about 4 hrs so not too bad

LW24Ai•8/31/24, 9:57 AM

Help me with Kohya_GUI_Flux_Installer_21, I have such a problem coming out, I have installed all the requirements of Python 3.10.11, FFmpeg, Cuda 11.8, C++ and Git tools as it was in the patrion post

LW24Ai•8/31/24, 10:13 AM

GGomezBro Anyone know what website this is? Since I have GPUS at home I never tried a 4 G...

Furkan Gözükara SECourses•8/31/24, 10:22 AM

massed compute

Furkan Gözükara SECourses•8/31/24, 10:22 AM

i made the tutorial editing but you can see here

Furkan Gözükara SECourses•8/31/24, 10:23 AM

https://youtu.be/bupRePUOA18

YouTubeSECourses

FLUX: The First Ever Open Source txt2img Model Truly Beats Midjourn...

FLUX is the first time every an open source txt2img is able to truly surpass and produce better quality and better prompt following images than #Midjourney, Adobe Firefly, Leonardo Ai, Playground Ai, Stable Diffusion, SDXL, SD3 and Dall E3. #FLUX is developed Black Forest Labs and its team is mainly composed by the original authors of #StableDi...

daroth•8/31/24, 10:23 AM

Is the Flux text encoder limited in token length?

Furkan Gözükara SECourses•8/31/24, 10:23 AM

14:35 How to use SwarmUI and FLUX on cloud services - no PC or GPU is required
14:48 How to use pre-installed SwarmUI on amazing Massed Compute 48 GB GPU for 31 cents per hour with FLUX dev FP16 model
16:05 How to download FLUX models on Massed Compute instance
17:15 FLUX models downloading speed of Massed Compute
18:19 How much time it takes on Massed Compute download all very best FP16 FLUX and T5 models

Ddaroth Is the Flux text encoder limited in token length?

Furkan Gözükara SECourses•8/31/24, 10:23 AM

512 i think

Oorpheus_reup My first lora using @Dr. Furkan Gözükara tutorial and config came out great. Us...

Furkan Gözükara SECourses•8/31/24, 10:23 AM

you can use both fp8 or fp16

daroth•8/31/24, 10:24 AM

ah its 77 for CLIP, 512 for T5

Ddaroth ah its 77 for CLIP, 512 for T5

Furkan Gözükara SECourses•8/31/24, 10:25 AM

yep

GGomezBro Couldnt I technically do like RANK 6 Batch 4 or 5 on 1 A6000? - Currently doing ...

Furkan Gözükara SECourses•8/31/24, 10:25 AM

over batch 2

Furkan Gözükara SECourses•8/31/24, 10:25 AM

doesnt bring any speed

Furkan Gözükara SECourses•8/31/24, 10:25 AM

batch 2 brings slight like 20%

LLW24Ai Click to see attachment

Furkan Gözükara SECourses•8/31/24, 10:26 AM

hello this is internet error

Furkan Gözükara SECourses•8/31/24, 10:26 AM

you can fix with warp vpn - cloudflrare

Furkan Gözükara SECourses•8/31/24, 10:26 AM

https://one.one.one.one/

1.1.1.1 — The free app that makes your Internet faster.

Install the free app that makes your phone’s Internet more fast, private, and reliable.

Furkan Gözükara SECourses•8/31/24, 10:26 AM

so activate warp vpn and do a fresh install

FFurkan Gözükara SECourses you can use both fp8 or fp16

orpheus_reupOP•8/31/24, 10:33 AM

Yes... but fp8 just not working with Lora. I don't know why. I just switch models, everything else stay the same.

Maybe I'll risk a forge update

FFurkan Gözükara SECourses https://one.one.one.one/

LW24Ai•8/31/24, 10:35 AM

The problem is not with the internet, I installed the original one and it works https://github.com/bmaltais/kohya_ss and your build is failing

GitHub

GitHub - bmaltais/kohya_ss

Contribute to bmaltais/kohya_ss development by creating an account on GitHub.

LLW24Ai The problem is not with the internet, I installed the original one and it works ...

Furkan Gözükara SECourses•8/31/24, 10:37 AM

You can see my install by editing file it is nothing special it installs that with accurate branch

Oorpheus_reup Yes... but fp8 just not working with Lora. I don't know why. I just switch model...

Furkan Gözükara SECourses•8/31/24, 10:38 AM

Ah could be forge thing

Furkan Gözükara SECourses•8/31/24, 10:38 AM

I used only Fp16 on forge

Sharinel•8/31/24, 10:42 AM

Just a quick FYI for anyone interested, here's how long it took me to do a 20img/150 epoch training session on a 4090 this morning. Used 'Rank 4 best config better colours' script

SSharinel Just a quick FYI for anyone interested, here's how long it took me to do a 20img...

Furkan Gözükara SECourses•8/31/24, 10:43 AM

Very nice speed

FFurkan Gözükara SECourses You can see my install by editing file it is nothing special it installs that wi...

LW24Ai•8/31/24, 10:50 AM

Is this error going to cause me problems down the road?

LLW24Ai Is this error going to cause me problems down the road?

Sharinel•8/31/24, 10:52 AM

https://discord.com/channels/772774097734074388/1248477696443416656/1279388443486716037

SSharinel https://discord.com/channels/772774097734074388/1248477696443416656/127938844348...

LW24Ai•8/31/24, 10:52 AM

Thanks

LLW24Ai Is this error going to cause me problems down the road?

Furkan Gözükara SECourses•8/31/24, 11:00 AM

Nope

Furkan Gözükara SECourses•8/31/24, 11:00 AM

We don't use xformers with FLUX

FFurkan Gözükara SECourses I used only Fp16 on forge 🙂

orpheus_reupOP•8/31/24, 11:02 AM

The fp16 works really well.

I used no captioning, no regs. Best Lora i ever made

Dragoy•8/31/24, 12:18 PM

@Dr. Furkan Gözükara Can you please tell me if this is considered a good learning curve for my 3090?

DDragoy @Dr. Furkan Gözükara Can you please tell me if this is considered a good learnin...

Furkan Gözükara SECourses•8/31/24, 12:19 PM

loss rate is meaningless

Furkan Gözükara SECourses•8/31/24, 12:19 PM

speed is looking ok not very bad

Furkan Gözükara SECourses•8/31/24, 12:19 PM

i think your model will be really good

Furkan Gözükara SECourses•8/31/24, 12:20 PM

trained a lora for a client i think it turned out to be perfect

Furkan Gözükara SECourses•8/31/24, 12:20 PM

can do smiling very well since we included that :d

mikemenders•8/31/24, 12:36 PM

While you guys are training Flux, I'm experiencing the interesting features of PonyRealism 2.1 under Prodigy. (I see that the Flux training is still being tweaked at Kohya, so I'll wait.) It turns out that the SDXL rule of only training Unet doesn't apply here, because then the finished Lora will only look a little like the person. However, if I train normally (i.e. without the --network_train_unet_only switch), I'm already on target. By using the 'fp8 base training' switch I can weaken the strength of the model a bit, i.e. sometimes it can come in handy in achieving a more accurate Lora. Without fp8, the Lora is stronger, so the d_coef switch needs to be reduced.

Mmikemenders While you guys are training Flux, I'm experiencing the interesting features of P...

Furkan Gözükara SECourses•8/31/24, 12:39 PM

for sdxl i suggest unet + text encoder 1

Furkan Gözükara SECourses•8/31/24, 12:39 PM

also for flux i am testing training clip large text encoder atm

dxqb•8/31/24, 12:48 PM

RealVisXL5 was released

Ddxqb RealVisXL5 was released

Furkan Gözükara SECourses•8/31/24, 12:49 PM

wow i should test

Furkan Gözükara SECourses•8/31/24, 12:50 PM

testing flux clip large trainings

FFurkan Gözükara SECourses wow i should test

dxqb•8/31/24, 12:59 PM

seems to be a big training effort, not just remerge: "Hello everyone! An early version of RealVisXL V5.0 (560k steps) is now available on Hugging Face. The model is still in training, and the final version will be released on Hugging Face and CivitAI at the end of August. You can leave your feedback on this version in the comments below."

GomezBro•8/31/24, 12:59 PM

@Dr. Furkan Gözükara Bro, I am going to sleep now its 6am haha, training loras all night, I fixed all my machines now. Anyway, I want to Train 1000 images (a style of art) I want to train that but doing it on 1 A6000 it would take like a week right? What system you recommend and how much would it cost me on Mass compute? 4-8 GPU A6000? If its $3-5 an HR I am down to pay it. Please do recommend the correct VM to use on there and do you have a video setting one up as I only use local, never rented hardware before. Let me know thanks and gn brother! Totally worth the Patreon fee!

Ddxqb seems to be a big training effort, not just remerge: "Hello everyone! An early v...

Furkan Gözükara SECourses•8/31/24, 1:27 PM

nice

GGomezBro @Dr. Furkan Gözükara Bro, I am going to sleep now its 6am haha, training loras a...

Furkan Gözükara SECourses•8/31/24, 1:28 PM

get 4x A6000 on massed compute

Furkan Gözükara SECourses•8/31/24, 1:28 PM

it is 1.25$ per hour with our coupon

Furkan Gözükara SECourses•8/31/24, 1:28 PM

i made the multi gpu training tutorial so it will be published today or tomorrow hopefully

My first lora using @Dr. Furkan Gözükara tutorial and config came out great. Using Forge it is work

Similar Threads