Software Engineering Courses (SECourses)•10mo ago

Running WAN on 3090 - 25 steps - 368x208 Yesterday I was getting 17-25 s/iteration. So about 3-5 mi

Running WAN on 3090 - 25 steps - 368x208
Yesterday I was getting 17-25 s/iteration. So about 3-5 mins to create.
Windows update happened last night, I am now getting 100-185s /iteration

Anyone seeing the same?

OOddjob Running WAN on 3090 - 25 steps - 368x208 Yesterday I was getting 17-25 s/iterati...

Furkan Gözükara SECourses•3/12/25, 4:39 PM

check how much ram and vram you are using

Furkan Gözükara SECourses•3/12/25, 4:39 PM

before starting

Furkan Gözükara SECourses•3/12/25, 4:40 PM

but doesnt make sense

nicearoni•3/12/25, 5:31 PM

not sure the update to the WAN app is working here to address the long LoRA loading time... I've got a 3090 and 64GB RAM, still seeing a really long wait at "Adding LoRA to wan_video_dit (..." step. how long should this take? is there anything i can check\change to improve?

nicearoni•3/12/25, 5:32 PM

even though i have 24GB VRAM i tried the 16GB profile but that did not seem to help

Versace Waterfall•3/12/25, 5:56 PM

Huge THANK YOU, got swarm working for the first time on my 5080. All my past attempts failed and I was only able to get comfy ui working.

nicearoni•3/12/25, 6:06 PM

I also tried changing torch type to torch.float8_e4m3fn and then it makes it past the 'Adding LoRA' step, but then after saving the auto pre processed image I get the error: RuntimeError: getCudnnDataTypeFromScalarType() not supported for Float8_e4m3fn

blackhat•3/12/25, 6:20 PM

https://www.patreon.com/posts/click-to-open-post-used-in-tutorial-110879657
@Dr. Furkan Gözükara i am implementing your traininig flow for person training can i get your training folder of the images your data set ? that also a text file as i can see in your vedio

Patreon

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

Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon

blackhat•3/12/25, 6:24 PM

@Dr. Furkan Gözükara i guess that is a file which have the config or the captioning of the images or somethin would be please to know if you can give me your 15 image dataset folder with the file

Bill Meeks•3/12/25, 6:33 PM

I just dropped a new model, the last piece of the puzzle for making Very Special, inspired by Saturday Morning Cartoons from the 80s and 90s. Enjoy! https://everlyheights.tv/tools/stable-diffusion-models/lora/everly-heights-saturday-morning-flux/

Ddxqb Flux with negative prompting

Dubey•3/12/25, 6:41 PM

hey @dxqbYD how do you do negative prompting with flux? I thought you can't do it

DDubey hey @dxqbYD how do you do negative prompting with flux? I thought you can't do i...

Dubey•3/12/25, 6:43 PM

I also saw your other pictures where you said you can get better outputs with different prompting.. how are you doing that?

VVersace Waterfall Huge THANK YOU, got swarm working for the first time on my 5080. All my past att...

Furkan Gözükara SECourses•3/12/25, 6:44 PM

awesome

Nnicearoni I also tried changing torch type to torch.float8_e4m3fn and then it makes it pas...

Furkan Gözükara SECourses•3/12/25, 6:45 PM

are you making complete fresh install

Furkan Gözükara SECourses•3/12/25, 6:45 PM

because i tested fp8 works both on rtx 3090 ti and 5090

Furkan Gözükara SECourses•3/12/25, 6:45 PM

did you also follow requirements video;?

Bblackhat https://www.patreon.com/posts/click-to-open-post-used-in-tutorial-110879657 @Dr....

Furkan Gözükara SECourses•3/12/25, 6:45 PM

our example dataset is here

Furkan Gözükara SECourses•3/12/25, 6:46 PM

https://www.patreon.com/posts/114972274

Furkan Gözükara SECourses•3/12/25, 6:46 PM

example model is here with that dataset https://civitai.com/models/911087/dwayne-johnson-aka-the-rock-flux-dev-fine-tuning-dreambooth-model-for-educational-and-research-purposes-dwayne-johnson-aka-the-rock-flux-dev-lora-model-for-educational-and-research-purposes-full-tutorial

Nnicearoni not sure the update to the WAN app is working here to address the long LoRA load...

Furkan Gözükara SECourses•3/12/25, 6:47 PM

if i send you a py file

Furkan Gözükara SECourses•3/12/25, 6:47 PM

can you replace it?

Furkan Gözükara SECourses•3/12/25, 6:47 PM

for debuggng purposes of speed

Furkan Gözükara SECourses•3/12/25, 6:48 PM

ok i will pm you so we can see what is taking time

nicearoni•3/12/25, 6:49 PM

i verified i have cuda 12.8 and have the latest working kohya and swarm. fp16 image2vid works, just takes a long time at the LoRA step.

DDubey hey @dxqbYD how do you do negative prompting with flux? I thought you can't do i...

blackhat•3/12/25, 6:50 PM

@Dr. Furkan Gözükara can you tell me something one this , like how he is able to do it with negative prompting

Nnicearoni i verified i have cuda 12.8 and have the latest working kohya and swarm. fp16 im...

Furkan Gözükara SECourses•3/12/25, 6:52 PM

ye for debugging reason

Bblackhat @Dr. Furkan Gözükara can you tell me something one this , like how he is able to...

Furkan Gözükara SECourses•3/12/25, 6:52 PM

use de distilled model

xlads•3/12/25, 6:52 PM

ERROR: Failed to restore /workspace/supir files/SUPIR/venv/bin/convert-caffe2-to-onnx
ERROR: Failed to restore /workspace/supir files/SUPIR/venv/bin/convert-onnx-to-caffe2
ERROR: Failed to restore /workspace/supir files/SUPIR/venv/bin/torchfrtrace
ERROR: Failed to restore /workspace/supir files/SUPIR/venv/bin/torchrun
ERROR: Could not install packages due to an OSError: [Errno 122] Disk quota exceeded

Furkan Gözükara SECourses•3/12/25, 6:52 PM

cfg 3.5 and negative prompt

Xxlads ERROR: Failed to restore /workspace/supir files/SUPIR/venv/bin/convert-caffe2-to...

Furkan Gözükara SECourses•3/12/25, 6:52 PM

your disk is out of space

Furkan Gözükara SECourses•3/12/25, 6:53 PM

delete pod and expand

xlads•3/12/25, 6:56 PM

so I delete pod then deploy a bigger pod under same storage

Nnicearoni i verified i have cuda 12.8 and have the latest working kohya and swarm. fp16 im...

Furkan Gözükara SECourses•3/12/25, 6:56 PM

sent you file

Furkan Gözükara SECourses•3/12/25, 6:57 PM

edit your network volume and expand

Furkan Gözükara SECourses•3/12/25, 6:57 PM

@Random#7680174

Furkan Gözükara SECourses•3/12/25, 6:57 PM

xlads•3/12/25, 6:57 PM

okay thanks

xlads•3/12/25, 7:27 PM

My workspace got messy so quickly

xlads•3/12/25, 7:51 PM

getting this error now Enter your choice (1-2): 2
Please select an option:
1. Start With Light Theme
2. Start With Dark Theme

Enter your choice (1-2): 1
Traceback (most recent call last):
File "/workspace/SUPIR/gradio_demo.py", line 14, in <module>
import einops

xlads•3/12/25, 7:57 PM

This is in trying to install supir

blackhat•3/12/25, 8:00 PM

@Dr. Furkan Gözükara why its using only one gpu if we have 4 here ?

Xxlads getting this error now Enter your choice (1-2): 2 Please select an option: 1. St...

Furkan Gözükara SECourses•3/12/25, 8:01 PM

rm -r supir

Furkan Gözükara SECourses•3/12/25, 8:01 PM

re install from fresh

Bblackhat @Dr. Furkan Gözükara why its using only one gpu if we have 4 here ?

Furkan Gözükara SECourses•3/12/25, 8:01 PM

which app?

FFurkan Gözükara SECourses which app?

blackhat•3/12/25, 8:02 PM

runpod

Bblackhat runpod

Furkan Gözükara SECourses•3/12/25, 8:02 PM

it is not an app

blackhat•3/12/25, 8:02 PM

nvitop

blackhat•3/12/25, 8:04 PM

in runpod i am using L40 with 4 gpu and with swernui person implimentation
https://www.youtube.com/watch?v=-uhL2nW7Ddw&ab_channel=SECourses
@Dr. Furkan Gözükara

YouTubeSECourses

Blazing Fast & Ultra Cheap FLUX LoRA Training on Massed Compute & R...

Unlock the power of FLUX LoRA training, even if you're short on GPUs or looking to boost speed and scale! This comprehensive guide takes you from novice to expert, showing you how to use Kohya GUI for creating top-notch FLUX LoRAs in the cloud. We'll cover everything: maximizing quality, optimizing speed, and finding the best deals. With our exc...

Bblackhat in runpod i am using L40 with 4 gpu and with swernui person implimentation http...

Furkan Gözükara SECourses•3/12/25, 8:07 PM

did you load 4x gpu config?

Furkan Gözükara SECourses•3/12/25, 8:07 PM

tell me your config name

Running WAN on 3090 - 25 steps - 368x208 Yesterday I was getting 17-25 s/iteration. So about 3-5 mi

Similar Threads

Similar Threads

Similar Threads