fine tuning will have better quality
fine tuning will have better quality
Fthis is lora
Fi will next research fine tuning
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dalthough Ive seen some train lora on flux with around 13 epochs
D11 to 13 seems the sweet spot
Dlooks pretty good to me alrteady Doc
DOneTrainer dev is making the claim that you can have finetuning quality when using Dora-Loras, that they have recently implemented and I think Kohya had for a while
DI've not tested it, just might be something interesting for you
DDora Loras hehe just rolls of the tongue
Dit's only called DoRA, I just added LoRA to make clear that it's just another type of lora
DAll we need is Sora Dora Loras
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Dpaper by Liu, Wang, Yin andChen
DarXiv.org
Among the widely used parameter-efficient fine-tuning (PEFT) methods, LoRA and its variants have gained considerable popularity because of avoiding additional inference costs. However, there still often exists an accuracy gap between these methods and full fine-tuning (FT). In this work, we first introduce a novel weight decomposition analysis t...
Dfuture of AI seems Chinese when you look at the authors of so many papers
Ccan you use flux loras in swarm ?
Fyep
Fdora is supposed to be
Fyep i am using
Fright now
F
Fsupir is chinese model
Djust installed some old hdd to have more space for .safetensors. damn this is slow. don't switch from NVMe to HDD 
CIt works on photo real images, because model replaces all photo real men with the trained data.
But it doesnt learn what 'ohwx' tag means. So try adding "Cartoon image of ohwx man as Aladdin" see if it still keeps some of the ohwx features
But it doesnt learn what 'ohwx' tag means. So try adding "Cartoon image of ohwx man as Aladdin" see if it still keeps some of the ohwx features
Fit is painful
Fgonna check in a moment
Flooks like man prompt is strongly overwritten
Fprobably we need reg images
Fye lora is too powerful it made model overfit
Freally hard:D
F
DJeff Goldblum vibes
Ftrue
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
Fit really struggles
Fstylized outputs
Ftime to start new 8 trainings :d
SDr Furkan i hope everything is ok , STILL NOT OPTION FOR 16 gigas VRAM GPUS?
Fyep i am messaging kohya
Fhere you can see
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...
Fit uses 18 gb
Fbut here try it
Flowest_vram.json5.2KB
Ftell me your step speed / shared vram suage
Cthats actually not too bad though, it atleast took some of your likeness, not sure whats going on with my snow white prompt
Fi am doıing more tests
�Have you tried setting Lora strength to like 1.4? I've noticed sometimes that's what can really force it with some loras
Fi am using 1.0. it is already too powerful
F
FRaw vs eyeglasses prompt - it is interesting that FLUX is learning me without eyeglasses when not prompted. This was never case with SDXL
�Perhaps due to captioning with the dataset?