Software Engineering Courses (SECourses)•17mo ago

I thought so as well, but apparently a venv is not a 100% virtual environment. It still uses the sys

I thought so as well, but apparently a venv is not a 100% virtual environment. It still uses the system as well :/ To my disappointment

KKallamamran I thought so as well, but apparently a venv is not a 100% virtual environment. I...

J2_Nedo•8/23/24, 9:13 AM

hmm i was planing to install it, so you ended up with conflicts?

J2_Nedo•8/23/24, 9:14 AM

yesterday i installed, Python, Torch, Cuda ect. everthing fresh and clean

Tacite•8/23/24, 9:16 AM

how do you upscale your fluix images?

KKallamamran Hey @Dr. Furkan Gözükara Do we have any script for collecting similar images? I ...

Furkan Gözükara SECourses•8/23/24, 9:16 AM

i have similarity script

Furkan Gözükara SECourses•8/23/24, 9:17 AM

1 sec looking it

Iit @Dr. Furkan Gözükara one trainer is working on flux

Furkan Gözükara SECourses•8/23/24, 9:17 AM

oh good news

TTacite how do you upscale your fluix images?

Furkan Gözükara SECourses•8/23/24, 9:25 AM

SUPIR is best

Furkan Gözükara SECourses•8/23/24, 9:25 AM

how much upscale you need? FLUX can already natively generate

KKallamamran Hey @Dr. Furkan Gözükara Do we have any script for collecting similar images? I ...

Furkan Gözükara SECourses•8/23/24, 9:25 AM

added here : https://www.patreon.com/posts/88391247

Patreon

SOTA Subject Cropper and Face Focused Image Resizer Scripts Do Bett...

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

FFurkan Gözükara SECourses how much upscale you need? FLUX can already natively generate

Tacite•8/23/24, 9:26 AM

I generate on my machine with RTX 4080 at 1024x1024 and I'd like to batch upscale them, I'll try SUPIR on runpod then, thanks

TTacite I generate on my machine with RTX 4080 at 1024x1024 and I'd like to batch upscal...

Furkan Gözükara SECourses•8/23/24, 9:26 AM

ye first find good settings you like then give batch

Furkan Gözükara SECourses•8/23/24, 9:26 AM

it works supir

Tacite•8/23/24, 9:28 AM

I used the X/Y/Z plot with S/R replace to generate a grid with 3 lora trained on kohyaa flux (50, 75, 100 epoch) and at different strength (0.75, 1, 1.25), the 100 epoch at 1.25 was surprisingly good, almost same as 150 epoch at 1.0 strength. I will use 150 at 1.0 now I think

TTacite I used the X/Y/Z plot with S/R replace to generate a grid with 3 lora trained on...

Furkan Gözükara SECourses•8/23/24, 9:29 AM

nice that is the way to go

FFurkan Gözükara SECourses added here : https://www.patreon.com/posts/88391247

KallamamranOP•8/23/24, 9:51 AM

Really?

KallamamranOP•8/23/24, 9:52 AM

How does it rank similarities?

KKallamamran How does it rank similarities?

Furkan Gözükara SECourses•8/23/24, 11:10 AM

https://poe.com/s/GNVIQpjRgOfGIm6TZoJP

look below script and explain me how to find 80% similar images

Claude-3.5-Sonnet: The script you've provided is designed to find duplicate and near-duplicate images using image hashing. To find images that are 80% similar, we need to adjust the script slightly. I

it•8/23/24, 11:24 AM

@Dr. Furkan Gözükara how long does it take you for 3000 steps on 1024?

Iit @Dr. Furkan Gözükara how long does it take you for 3000 steps on 1024?

Furkan Gözükara SECourses•8/23/24, 11:25 AM

it depends on gpu

Furkan Gözükara SECourses•8/23/24, 11:25 AM

currently rtx 3090 is like 7.7 second / it

Furkan Gözükara SECourses•8/23/24, 11:25 AM

but on linux way faster

Furkan Gözükara SECourses•8/23/24, 11:25 AM

trying to fix this issue

it•8/23/24, 11:32 AM

tried using wsl?

Iit tried using wsl?

Furkan Gözükara SECourses•8/23/24, 11:32 AM

i didnt

Furkan Gözükara SECourses•8/23/24, 11:32 AM

i dont want to add extra hardness :d

Furkan Gözükara SECourses•8/23/24, 11:35 AM

i will ask pytorch github :d

Furkan Gözükara SECourses•8/23/24, 11:44 AM

https://github.com/pytorch/pytorch/issues/134324

GitHub

Kohya SS FLUX LoRA training is way faster on Linux than Windows any...

Describe the bug I am using Kohya SS to train FLUX LoRA On Linux RTX 3090 gets like 5.5 second / it - batch size 1 and 1024x1024 px resolution On Windows RTX 3090 TI gets 7.7 second / it - has the ...