Any news for Onetrainer doc?
Any news for Onetrainer doc?
Fi dont see any yet
Fthey dont even have branch on github
Tonetrainer is what has given me the best results with the possibility of applying the masks, that is not possible in kohya, right?
Fkohya has masks but i dont know how exactly it works
Fi also think onetrainer works better since 1 developer developing gui
Fwith kohya a seperate developer developing gui
Fso there is sometimes issues and dispercancy
V@Dr. Furkan Gözükara your model must know smaller token
so you need to have that data in training
like
small bear, yes I do not have that, I need that, how to do?
so you need to have that data in training
like
small bear, yes I do not have that, I need that, how to do?
V@Dr. Furkan Gözükara waiting for your reply, could you share me with the parameters?
Fcan you eloborate your question more?
Fif you are using small after training and still not generating small that means your base model and training dataset not good enough
Vi never use small
Vyou mean during the time tagging?
Vi use sdxl as the base model, and training data, do you mean if i need the panda to look smaller, i had better offer some pandas and people besides it, to show the proportion between them, one is smaller, one is bigger?
Fyes you need in your training dataset
For the base model you use must know it
Fnew script AuraSR GigaGAN 4x published
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözükara on Patreon
BIntroducing the Everly Heights Prompt Extractor - A Windows utility that extracts your original prompts from Stable Diffusion images for training. https://everlyheights.tv/tools/stable-diffusion-tools/everly-heights-prompt-extractor-for-stable-diffusion/
Fyou should publish source of exe on github
Fotherwise i wouldnt trust any exe
BThe GitHub is linked to on the product page. It's an optional donation, or just click the button and pull it down free from GitHub.
BSource included.
BGitHub
Contribute to MrBillMeeks/Everly-Heights-Prompt-Extractor development by creating an account on GitHub.
Fnew comparison : https://www.reddit.com/r/StableDiffusion/comments/1esb128/aurasr_gigagan_4x_upscaler_testing_works_amazing/
Reddit
Explore this post and more from the StableDiffusion community
Fthat is why i am still waiting kohya to finalize. otherwise tutorial and trainings becomes too soon obsolete
https://www.reddit.com/r/StableDiffusion/comments/1es91bu/major_bug_affecting_all_flux_training_and_causing/
https://www.reddit.com/r/StableDiffusion/comments/1es91bu/major_bug_affecting_all_flux_training_and_causing/
Reddit
Explore this post and more from the StableDiffusion community
WI am loading the SwarmUI and noticed this in the debug logs. It's taking several minutes for the model to load:
W2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
WHow do I keep it from casting it to bfloat16?
Flet me ask this
WIt seems like its loading into main system memory first, even though I have a 4090 which can easily hold it. Nvitop shows system memory usage steadily climbing
Wthis is using the dev model as well.
Fye i think it is designed that way
WOk, but it's been over 9 minutes and the model still hasn't loaded.
W2024-08-14 15:14:17.991 [Info] User local requested 1 image with model 'flux1-dev-fp8.safetensors'...
2024-08-14 15:14:17.997 [Debug] [BackendHandler] Backend request #1 for model flux1-dev-fp8.safetensors, maxWait=7.00:00:00.
2024-08-14 15:14:18.002 [Debug] [BackendHandler] backend #0 will load a model: flux1-dev-fp8.safetensors, with 1 requests waiting for 0 seconds
2024-08-14 15:14:20.374 [Debug] [ComfyUI-0/STDERR] got prompt
2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
2024-08-14 15:14:22.604 [Debug] [ComfyUI-0/STDERR] model_type FLUX
2024-08-14 15:14:17.997 [Debug] [BackendHandler] Backend request #1 for model flux1-dev-fp8.safetensors, maxWait=7.00:00:00.
2024-08-14 15:14:18.002 [Debug] [BackendHandler] backend #0 will load a model: flux1-dev-fp8.safetensors, with 1 requests waiting for 0 seconds
2024-08-14 15:14:20.374 [Debug] [ComfyUI-0/STDERR] got prompt
2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
2024-08-14 15:14:22.604 [Debug] [ComfyUI-0/STDERR] model_type FLUX
WAt this point it's 13 minutes
Fsomething seems wrong
Fhow much RAM you have?
W24gb, 64gb of system ram
Fok both good
Whold on it finally returned let me get the log
Fcan you check your hdd reading speed
Wmessage.txt4.93KB
Fi think your hdd was too slow
Whmm.. ok I'll move it to a faster drive and see if it changes.
Fsure give it a try
Fi am also adding new downloads to downloader
F4bit bnb
Fmerged model