Software Engineering Courses (SECourses)•10mo ago

Hi @Dr. Furkan Gözükara, I noticed in the WAN 2.1 14B Image-to-Video 720P specifications - it shows

Hi @Dr. Furkan Gözükara,
I noticed in the WAN 2.1 14B Image-to-Video 720P specifications - it shows "Number of Persistent Parameters In Dit (VRAM)" as 0 for 24GB cards. Is this correct, or possibly an error?
I'm asking because I'm observing my VRAM being fully utilized despite the parameter count showing as 0.

part.time.artist•3/2/25, 1:07 PM

I've scanned this thread but I am getting an error when running V23 of the model downloader:

Enter your choices (comma or dash separated, e.g., 1,3,5 or 1-3-5): 4
Created or verified directory: SwarmUI\Models
Created or verified directory: SwarmUI\Models\vae
Created or verified directory: SwarmUI\Models\diffusion_models
Created or verified directory: SwarmUI\Models\Stable-Diffusion
Created or verified directory: SwarmUI\Models\clip
Created or verified directory: SwarmUI\Models\yolov8
Created or verified directory: SwarmUI\Models\style_models
Created or verified directory: SwarmUI\Models\Lora
Created or verified directory: SwarmUI\Models\upscale_models
Traceback (most recent call last):
File "C:\Projects\SwarmUI\Download_Models.py", line 328, in <module>
download_models(selected_options)
File "C:\Projects\SwarmUI\Download_Models.py", line 66, in download_models
snapshot_download(
File "C:\Users\john\AppData\Local\Programs\Python\Python310\lib\site-packages\huggingface_hub\utils_deprecation.py", line 101, in inner_f
return f(args, **kwargs)
File "C:\Users\john\AppData\Local\Programs\Python\Python310\lib\site-packages\huggingface_hub\utils_validators.py", line 124, in _inner_fn
return fn(args, **kwargs)
TypeError: snapshot_download() got an unexpected keyword argument 'local_dir'

Ppart.time.artist I've scanned this thread but I am getting an error when running V23 of the model...

part.time.artist•3/2/25, 1:14 PM

I resolved my issue. I opened the python file and manually ran the pip installs at the top and it seems to be downloading

Furkan Gözükara SECourses•3/2/25, 3:06 PM

guys i got a baby boy today so i may reply later today hopefully

ZZaulinator i just see merged wan models on civitai, so just curious if we would be able to ...

Furkan Gözükara SECourses•3/2/25, 3:06 PM

yes hopefully i will add lora asap

KKarandeep Singh Hi @Dr. Furkan Gözükara, I noticed in the WAN 2.1 14B Image-to-Video 720P specif...

Furkan Gözükara SECourses•3/2/25, 3:07 PM

yes it uses more therefore 0

Furkan Gözükara SECourses•3/2/25, 3:07 PM

and still uses your entire vram

Furkan Gözükara SECourses•3/2/25, 3:07 PM

you can try fp8

Ppart.time.artist I resolved my issue. I opened the python file and manually ran the pip installs ...

Furkan Gözükara SECourses•3/2/25, 3:07 PM

yes i was gonna tell this

Furkan Gözükara SECourses•3/2/25, 3:07 PM

library version issue

Furkan Gözükara SECourses•3/2/25, 3:07 PM

i am glad you solved

FFurkan Gözükara SECourses guys i got a baby boy today so i may reply later today hopefully

part.time.artist•3/2/25, 3:16 PM

Congratulations!

Ppart.time.artist Congratulations!

Furkan Gözükara SECourses•3/2/25, 3:17 PM

thanks

Furkan Gözükara SECourses•3/2/25, 3:17 PM

i got to be afk again

blackhat•3/2/25, 3:17 PM

@Dr. Furkan Gözükara can you help me in figuring out the right data set for the person lora to get full body features right like you have done with your 250 images i want to know about the data set and the prompts you set for it in flux

FFurkan Gözükara SECourses guys i got a baby boy today so i may reply later today hopefully

blackhat•3/2/25, 3:18 PM

congratulations

part.time.artist•3/2/25, 3:18 PM

I have an older RTX 2070 super, 8GB VRAM

I am manually installing Flux Schnell, and I've tried to watch as many of the videos and read the patreon posts to sed if anything you've covered would run on my hardware or not. I know some of the video models say "less than X VRAM" but are there any that could actually run on 8GB VRAM?

Joshua A•3/2/25, 5:52 PM

@Dr. Furkan Gözükara Have you done any Flux Lora tests on multi GPU ie 2 A100? I am looking to find a good number of steps and learning rate. It's limited to only 10 images. no more or less for training.

AIGambino•3/2/25, 6:44 PM

AIGambino•3/2/25, 6:44 PM

installing swarm from patreon, is this normal dr?

Timson•3/2/25, 7:06 PM

Avast :kek:
People still use it in this decade?

Timson•3/2/25, 7:07 PM

last time I heard of it was like, 2006

Bblackhat congratulations

Furkan Gözükara SECourses•3/2/25, 7:24 PM

thanks

AAIGambino Click to see attachment

Furkan Gözükara SECourses•3/2/25, 7:25 PM

yes

Bblackhat @Dr. Furkan Gözükara can you help me in figuring out the right data set for the ...

Furkan Gözükara SECourses•3/2/25, 7:25 PM

right dataset is having a lot of mix of shots

Furkan Gözükara SECourses•3/2/25, 7:25 PM

full body shot

Furkan Gözükara SECourses•3/2/25, 7:25 PM

close shot

Furkan Gözükara SECourses•3/2/25, 7:25 PM

half body shot

Furkan Gözükara SECourses•3/2/25, 7:25 PM

different clothing and backgrounds help too

JJoshua A @Dr. Furkan Gözükara Have you done any Flux Lora tests on multi GPU ie 2 A100? I...

Furkan Gözükara SECourses•3/2/25, 7:25 PM

well i did test for a client for 5 min training

Furkan Gözükara SECourses•3/2/25, 7:26 PM

if you want faster you need to increase batch size + learning rate

TTimson Avast :kek: People still use it in this decade?

Furkan Gözükara SECourses•3/2/25, 7:26 PM

yes avast itself is like a virus :d

FFurkan Gözükara SECourses right dataset is having a lot of mix of shots

blackhat•3/2/25, 7:26 PM

how big it need to be and how much variation of each is needed if you can mention a bit

blackhat•3/2/25, 7:28 PM

can i see yours models data set if possible and the traingin flow if possible

Bblackhat how big it need to be and how much variation of each is needed if you can mentio...

Furkan Gözükara SECourses•3/2/25, 7:29 PM

i shared it actually on patreon

1nfern0•3/2/25, 7:29 PM

just catching up on the weekend's developments. Excellent improvements.

Furkan Gözükara SECourses•3/2/25, 7:29 PM

11nfern0 just catching up on the weekend's developments. Excellent improvements.

Furkan Gözükara SECourses•3/2/25, 7:29 PM

thanks

Furkan Gözükara SECourses•3/2/25, 7:29 PM

hopefully gonna add lora next

Furkan Gözükara SECourses•3/2/25, 7:29 PM

also speed up coming soon

Furkan Gözükara SECourses•3/2/25, 7:29 PM

i am expecting teacache

1nfern0•3/2/25, 7:29 PM

I did get an error when testing RIFE. Should I just instal ffmpeg separately

11nfern0 I did get an error when testing RIFE. Should I just instal ffmpeg separately

Furkan Gözükara SECourses•3/2/25, 7:32 PM

yes you have to

Furkan Gözükara SECourses•3/2/25, 7:32 PM

this tutorial is really mandatory https://youtu.be/DrhUHnYfwC0

YouTubeSECourses

How to Install Python, CUDA, cuDNN, C++ Build Tools, FFMPEG & Git T...

If you are interested in using AI, generative AI applications, open source applications in your computer, than this is the most fundamental and important tutorial that you need. In this tutorial I show and explain how to properly install appropriate Python versions accurately, how to switch between different Python versions, how to install diffe...

FFurkan Gözükara SECourses well i did test for a client for 5 min training

Joshua A•3/2/25, 7:34 PM

It's not so much about the speed as the increasing the quality. I and reduce overfitting. If you were limited to only 10 images to train a loRA. What would you set the learning rate and number of steps?

right now we are at 4-e5 and 1600 steps. But this is on one a40 GPU.

JJoshua A It's not so much about the speed as the increasing the quality. I and reduce ove...

Furkan Gözükara SECourses•3/2/25, 7:34 PM

you need to test

Furkan Gözükara SECourses•3/2/25, 7:34 PM

you need to target a step count

Furkan Gözükara SECourses•3/2/25, 7:34 PM

and test learning rates for that

Joshua A•3/2/25, 7:36 PM

Ok. So first dial in step count. I wasn't sure if you had any experience before I start testing.

JJoshua A Ok. So first dial in step count. I wasn't sure if you had any experience before ...

Furkan Gözükara SECourses•3/2/25, 7:47 PM

nope

FFurkan Gözükara SECourses Click to see attachment

blackhat•3/2/25, 7:52 PM

in which article https://www.patreon.com/posts/111891669 is this one or something else

Patreon

Training a FLUX LoRA with 256 Images Experiments - Full Workflow an...

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

Hi @Dr. Furkan Gözükara, I noticed in the WAN 2.1 14B Image-to-Video 720P specifications - it shows

Similar Threads

Similar Threads

Similar Threads