add some blocks to swap, will reduce vram
add some blocks to swap, will reduce vram
X
AHi Guys... it's black-friday/cybermonday time... do you have any advice on what to buy to make flux/sdxl/pdxl models locally? I mean: what minimum requirements for a graphic card that can handle those kind of jobs... something cheap, as I do training only to understand, not for selling my works around, so I can bear also longest trainings... woud appreciate model cards or laptops adequately equipped advices, thanks to all!
Fawesome
Flooks to me like burnt
Frtx 3090 is min
Fi recommend
Fit is 1130 usd here right now
Fi just had to purchase kaspersky vpn - discord blocked cloudflare zero trust vpn ips 
Xif you want to train flux specifically, and also sdxl and other models you can on onetrainer . i think if you're getting started and on a tight budget, you can optimize many vram options and get a 4060ti 16gb
Xits going for 439.99 usd
Xcan also find deals for 399.99
Fyes this would also work
Fbut if something comes next with absolute minimum 24 gb 
Xtrue i saw 3090 on amazon for 1049 usd
Xmany good deals out
MHello when were training SDXL lora should we choose save precision float or bf16?
MI know we choose it float in flux
Fsave as float or fp16
Falmost same quality
Fi didnt test bf16
ZHello Everyone! What should be written in the "class prompt" field if you're training the style or aesthetic of Dark Fantasy 80s films? @Dr. Furkan Gözükara does not mention this in any of the tutorials, only using the word "a man."..
XDark Fantasy 80s films Style @Poly
Xstyle is your class prompt in this instance
ZThank you for the advice! Should I create captions for the style?
Xfor styles yes, you probably will get better results using tags for style type training
Xwd14
Min this video at 15:22 both mixed and save precision are set to bf16 https://youtu.be/-xEwaQ54DI4?si=OSy9G978Bpu1QbZy&t=920
YouTubeSECourses
If you don't have a GPU, or have a strong GPU, or you are using Mac and your computer not supporting Stable Diffusion training, SDXL training, then this is the tutorial you are looking for. In this tutorial, we will use a cheap cloud GPU service provider RunPod to use both Stable Diffusion Web UI Automatic1111 and Stable Diffusion trainer Kohya ...
Ftry prompts and see which one can generate most accurate one on flux dev
Fand pick that
Fit was sdxl
Ffor sdxl yes use those
BHow are steps calculated per epoch in Flux training?
Fimage count * repeating * epoch / batch size / gradient accumulation steps
AWhat causes this?
AHi, I have a problem with kohya - I was using the installer but these files are missing: /workspace/transformer/diffusion_pytorch_model-00001-of-00003.safetensors
A2 and 3 as well. Is there a way to download these by a command?
AI am using Kohya_FLUX_DreamBooth_v12.zip
AUsing runpod
Fover training
Fwhich base model did you given?
Fsave your json and private message me
Fit
AThank you 
AAny things I should change?
AHmm, any chance to change the save every n epochs while the training is running?
Fi dont remember exactly config but you shouldnt
Fdid you compare checkpoints?
Fno i think
Fbut editing toml would work or not not sure