i plan to add lora support
i plan to add lora support
Zi just see merged wan models on civitai, so just curious if we would be able to use those
EIs multi resolution training adviseable in Kohya? I see there's an "enable_buckets" setting - does it improve quality?
KHi @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.
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.
PI'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'
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'
PI resolved my issue. I opened the python file and manually ran the pip installs at the top and it seems to be downloading
Fguys i got a baby boy today so i may reply later today hopefully
Fyes hopefully i will add lora asap
Fyes it uses more therefore 0 
Fand still uses your entire vram
Fyou can try fp8
Fyes i was gonna tell this
Flibrary version issue
Fi am glad you solved
PCongratulations!
Fthanks
Fi got to be afk again
B@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
Bcongratulations
PI 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?
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?
J@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.
A
Ainstalling swarm from patreon, is this normal dr?
TAvast :kek:
People still use it in this decade?
People still use it in this decade?
Tlast time I heard of it was like, 2006
Fthanks
Fyes
Fright dataset is having a lot of mix of shots
Ffull body shot
Fclose shot
Fhalf body shot
Fdifferent clothing and backgrounds help too
Fwell i did test for a client for 5 min training
Fif you want faster you need to increase batch size + learning rate
Fyes avast itself is like a virus :d
Bhow big it need to be and how much variation of each is needed if you can mention a bit
Bcan i see yours models data set if possible and the traingin flow if possible
Fi shared it actually on patreon
1just catching up on the weekend's developments. Excellent improvements.
F
Fthanks
Fhopefully gonna add lora next
Falso speed up coming soon
Fi am expecting teacache
1I did get an error when testing RIFE. Should I just instal ffmpeg separately
Fyes you have to
Fthis tutorial is really mandatory https://youtu.be/DrhUHnYfwC0
YouTubeSECourses
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...
JIt'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.
right now we are at 4-e5 and 1600 steps. But this is on one a40 GPU.
Fyou need to test
Fyou need to target a step count