Software Engineering Courses (SECourses)•17mo ago

5.65 So it's expected for each Lora training to take somewhere around an hour and a half to two hour

5.65
So it's expected for each Lora training to take somewhere around an hour and a half to two hours?

Ddaryhead 5.65 So it's expected for each Lora training to take somewhere around an hour an...

Furkan Gözükara SECourses•9/2/24, 11:59 AM

on that gpu yes

Furkan Gözükara SECourses•9/2/24, 11:59 AM

rtx 4090 faster than 2x :d

FFurkan Gözükara SECourses https://youtu.be/nySGu12Y05k

ghstoic•9/2/24, 11:59 AM

Thanks, I followed it and restarted PC but same error still persists. Surprised it could be a RAM issue as I didn't get this issue before and am trying Rank_4 config (17gb) on a RTX 3090 with 24gb vram

Furkan Gözükara SECourses•9/2/24, 11:59 AM

you can also rent multiple

Gghstoic Thanks, I followed it and restarted PC but same error still persists. Surprised ...

Furkan Gözükara SECourses•9/2/24, 11:59 AM

can you send me your json file from private mesage

ghstoic•9/2/24, 12:00 PM

sure

Charkel•9/2/24, 12:17 PM

I'm trying to run on kaggle but i get an error with the backend after I am doing the cancel run and restarting

CCharkel I'm trying to run on kaggle but i get an error with the backend after I am doing...

Furkan Gözükara SECourses•9/2/24, 12:18 PM

yes kaggle broken

Charkel•9/2/24, 12:18 PM

Furkan Gözükara SECourses•9/2/24, 12:18 PM

here : https://github.com/pytorch/pytorch/issues/134929

GitHub

On Kaggle : libcusparse.so.12: undefined symbol: __nvJitLinkComplet...

Describe the bug I have tried everything but no luck Waiting your inputs to try more I tried torch 2.4.0, 2.5 - dev, cu 118, cu121 and cu124 - all same error This below code - I got the same erro...

On Kaggle : libcusparse.so.12: undefined symbol: __nvJitLinkComplet...

Furkan Gözükara SECourses•9/2/24, 12:19 PM

it is because of comfyui upgrade

Furkan Gözükara SECourses•9/2/24, 12:19 PM

i still couldnt find a solution opened an issue thread

Furkan Gözükara SECourses•9/2/24, 12:19 PM

i guess we have to wait kaggle to upgrade torch to 2.4

FFurkan Gözükara SECourses on that gpu yes

daryheadOP•9/2/24, 12:19 PM

How fast is the fastest single gpu?

Charkel•9/2/24, 12:19 PM

Ok thank you for confirming

Ddaryhead How fast is the fastest single gpu?

Furkan Gözükara SECourses•9/2/24, 12:20 PM

single fastest gpu is 4090 = L40S = H100 - on each one having different vRAM

Furkan Gözükara SECourses•9/2/24, 12:20 PM

it is like more than 2x fast than rtx 3090

Furkan Gözükara SECourses•9/2/24, 12:34 PM

testing best quality ranking is hard

Furkan Gözükara SECourses•9/2/24, 12:34 PM

the images are so close

AAfroman4peace I have created a lora that was based on the combined output of my sdxl model+ my...

Sipriyani•9/2/24, 12:49 PM

I have read someone used, fine tuned/ trained sdxl synthetic images of him self, to train flux lora. I don't know what exactly this benefits tho. I agree it may create a better own dataset img source, instead of an iphone picture, infact we don't use reg images in lora.

Furkan Gözükara SECourses•9/2/24, 1:37 PM

newest findings

Furkan Gözükara SECourses•9/2/24, 1:37 PM

Text encoder on 832px resolution yields better results than without text encoder 896px, thus you can reduce resolution and keep text encoder training on - Clip-L

896px without T5 attention mask is better than, 768px with T5 attention mask on

sergio_gonzo•9/2/24, 1:55 PM

Dr Furkan i hope verything is ok What would be the equivalent of an avatar D-ID that would allow me to chat with the avatar that uses chatgpt and gesture its responses in real time?

Hastur•9/2/24, 2:06 PM

Hi, I would like to test your prompts on ComfyUI. Do you know how to implement this part : "<segment:face,0.7>photo of ohwx man" in ComfyUI?

Ssergio_gonzo Dr Furkan i hope verything is ok What would be the equivalent of an avatar D-I...

Furkan Gözükara SECourses•9/2/24, 2:08 PM

live portrait is best you can find

Furkan Gözükara SECourses•9/2/24, 2:09 PM

but it is not audio to animation

HHastur Hi, I would like to test your prompts on ComfyUI. Do you know how to implement t...

Furkan Gözükara SECourses•9/2/24, 2:09 PM

sadly i dont know can you ask to comfyui discord?

Furkan Gözükara SECourses•9/2/24, 2:09 PM

https://discord.gg/YxWQmsmu

Hastur•9/2/24, 2:09 PM

Sure, that would be good to know

LW24Ai•9/2/24, 2:40 PM

There's no resemblance at all. Lora trained on 15 images for 8 hours((((

LW24Ai•9/2/24, 2:41 PM

I guess my Flux model doesn't fit Lora. My forge only supports fp8 models

LLW24Ai There's no resemblance at all. Lora trained on 15 images for 8 hours((((

Furkan Gözükara SECourses•9/2/24, 2:42 PM

try in swarmui

Furkan Gözükara SECourses•9/2/24, 2:42 PM

if you did accurate settings it should work

FFurkan Gözükara SECourses try in swarmui

LW24Ai•9/2/24, 2:47 PM

Does it work locally or only with cloud support?

LLW24Ai Does it work locally or only with cloud support?

Furkan Gözükara SECourses•9/2/24, 2:47 PM

both

Furkan Gözükara SECourses•9/2/24, 2:47 PM

wndows main tutorial

Furkan Gözükara SECourses•9/2/24, 2:47 PM

https://youtu.be/HKX8_F1Er_w

YouTubeSECourses

Zero to Hero Stable Diffusion 3 Tutorial with Amazing SwarmUI SD We...

Do not skip any part of this tutorial to master how to use Stable Diffusion 3 (SD3) with the most advanced generative AI open source APP SwarmUI. Automatic1111 SD Web UI or Fooocus are not supporting the #SD3 yet. Therefore, I am starting to make tutorials for SwarmUI as well. #StableSwarmUI is officially developed by the StabilityAI and your m...

Furkan Gözükara SECourses•9/2/24, 2:47 PM

windows flux tutorial 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...

Furkan Gözükara SECourses•9/2/24, 2:57 PM

newest ranks added https://www.patreon.com/posts/110879657

Patreon

Kohya FLUX LoRA Training Full Tutorial For Local Windows and Cloud ...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

Neuromasters•9/2/24, 3:21 PM

I have a generation machine in my office and the new forge does not work gradio no '''custom_component''' anyone solved the problem?

LLW24Ai There's no resemblance at all. Lora trained on 15 images for 8 hours((((

TheToday•9/2/24, 3:40 PM

are you using faceswap on the training images?

LLW24Ai There's no resemblance at all. Lora trained on 15 images for 8 hours((((

Tacite•9/2/24, 3:45 PM

Use higher quality and resolution images for your dataset

FFurkan Gözükara SECourses sadly i dont know can you ask to comfyui discord?

Tacite•9/2/24, 3:47 PM

<segment:face,0.7> si from SwarmUI ?

TTacite Use higher quality and resolution images for your dataset

LW24Ai•9/2/24, 3:47 PM

I used a 1024x1024 image with good quality images

TTheToday are you using faceswap on the training images?

LW24Ai•9/2/24, 3:47 PM

LLW24Ai I used a 1024x1024 image with good quality images

Tacite•9/2/24, 3:47 PM

the second image (on the right) is from your dataset? because that's low quality to me

TTacite <segment:face,0.7> si from SwarmUI ?

Furkan Gözükara SECourses•9/2/24, 3:48 PM

yes

TTacite the second image (on the right) is from your dataset? because that's low quality...

LW24Ai•9/2/24, 3:48 PM

How good do the images have to be? They were taken with an iphone camera.

LW24Ai•9/2/24, 3:50 PM

Original image size 1080x1350

Tacite•9/2/24, 3:51 PM

I can see heavy JPEG compression on that image, lighting is good but there's hardly any "texture". Maybe try the RAW MAX mode if you have it on your iphone, I used that for my dataset I have an iPhone 15 Pro

5.65 So it's expected for each Lora training to take somewhere around an hour and a half to two hour

Similar Threads

Similar Threads

Similar Threads