Yeah but the setting, ill test it, where is it
Yeah but the setting, ill test it, where is it
Gi can try it once i reinstall windows
Fwhich settings?
Fbatch size is on gradio
Fi got to sleep
Gill check it and increase it to 4
Gand see if it works
ADoes anyone know if ai toolkit you need to be connected to the internet for it to work?
BThe final slate of characters for my animated pilot Very Special. LoRAs trained using datasets built with the Everly Heights Character Maker.
Pi am rejected acces to Lama on HF. if i use your 1 click installer will i still need to acces HF ? or will it just install?
Fit will install you dont need
Fmy installers handle that 
Fwow new training parameter made huge difference
resemblance reduced but clothing much better
looks like i have to investigate
resemblance reduced but clothing much better
looks like i have to investigate
Fcheck this messagfe
FGitHub
Replace SD3Tokenizer with the original CLIP-L/G/T5 tokenizers.
Extend the max token length to 256 for T5XXL.
Refactor caching for latents.
Refactor caching for Text Encoder outputs
Extract arch...
Extend the max token length to 256 for T5XXL.
Refactor caching for latents.
Refactor caching for Text Encoder outputs
Extract arch...
Ffor stylized outputs it has a huge impact
Fi will do 8 more trainings with it
F
Mtop left looks accurate
F
Nis there explanation for what these flags do in the JoyCaption app? none of them are checked in any of the sample screenshots?
Fye you dont need to check them
Fthey are extra features for expert i didnt find tghem better :d
Fdefualt settings are best i found
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Ddoes it still cut off the caption
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fyes but i added option so script removes that part
S@Dr. Furkan Gözükara What do you think of possible training of clip_l for Flux lora training? Do you think it could improve the training even more?
Fkohya just announced that
FFi will test it hopefully
SHahah yeah I saw that, but I am curious how the results will be
Fi am updating configs oın patroen right now
F
Fstill calculating vrams
SAwesomee
�Interesting article about Flux LoRA training experiments with captions
https://civitai.com/articles/6982
https://civitai.com/articles/6982
The last few days were pretty shit in terms of burning money for GPU compute and getting nothing in return for it, haha. My NSFW model hit its limi...
F8 gb config added
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
F
Fi will add newest experiments too hopefully
Tgood job Dr
Fty
FHello, I tried massedcompute but after a 4h lora training I couldn't access the Running Instance. So I'm going back to Runpod, I'm going to use cloud hosting because I don't want to restart a gpu each time and start from scratch.
My question is, does anyone already do this? If so, is there an estimate of the cost over a month, for example?
My question is, does anyone already do this? If so, is there an estimate of the cost over a month, for example?
Fwhat you mean you couldnt?
Fi have been using for months never had such issue
F@NicB can you check
NCould you dm me your account email? happy to take a look
NOh i think if i read this correctly you are looking for persistent storage? so you don't have to re-install everything each time?
Nit wasn't that the instance was broken or had issues correct?
FI'm on a Mac, I can't use ThinLinc but Chrome, and every time the page loops, I tried to connect to it for over an hour after the end of training without success, so I turned off the GPU.