does Gradient accumulate steps make lora quality bad?
does Gradient accumulate steps make lora quality bad?
�if all training pics are 2048x2048, should I keep them that way, or reduce to 1024x1024?
RHow do I install xformers
Fit is increasing batch size
Fdont do it unless you know what you are doing
R
FYouTubeSECourses
Welcome to the Pillars of AI Application Libraries: How to Install Tutorial! 
Are you ready to embark on a journey of installing AI libraries and applications with ease? 
In this video, we'll guide you through the process of installing Python, Git, Visual Studio C++ Compile tools, and FFmpeg on your Windows 10 machine. 
We'll also show you ho...
Are you ready to embark on a journey of installing AI libraries and applications with ease? In this video, we'll guide you through the process of installing Python, Git, Visual Studio C++ Compile tools, and FFmpeg on your Windows 10 machine. We'll also show you ho...Rokay i look at that video
Xyeah that's what I want to do, since 12GB vram cant handle 2batch size with prodigy I was gonna add some accu steps, but saw some people say that it does lower the quality
Fbigger batch size = lowers quality
Ftrue
Rdo i need s pefic version of torch or would the most recent work
ROkay xformers is installing now, thanks
Rdo i use cuda 11.8 or 12.4
Ryour video used 11.8
KI have a general question for anyone who might know: When working with existing Flux checkpoints, should I:
Take an existing checkpoint and train my concepts on top of it, or
Merge my checkpoint (containing my concepts) with another existing checkpoint?
Will both approaches yield the same results? I'm wondering if training on top of an existing checkpoint is better, or if it's just a waste of time when I could simply merge checkpoints instead
Take an existing checkpoint and train my concepts on top of it, or
Merge my checkpoint (containing my concepts) with another existing checkpoint?
Will both approaches yield the same results? I'm wondering if training on top of an existing checkpoint is better, or if it's just a waste of time when I could simply merge checkpoints instead
Fcuda 12.4
Fnow we recommend that
Ftrain top of it
F= best results
Rbut nothing works if in use 12.4
Fmy newer scripts are all working
Fyou have to modify accordignly what you use
Ri tried, itry again
RRokay bitsandbyes can't install
R
Fye these are hardest things to solve
AOne question concerning epochs with dreambooth - if you have 10 images / medium quality, how many epochs with 1 repeats do you recommend?
Fi would go up to 300
Fand compare
Fbut 10 is low
Rdo you think there is a conflict bewwteen 12.4 and 11.8
AI had the best result at 400 - sadly these are very old family photos, so not enough available photos of my parents 
AI think after 400 or 500 it starts to overtrain.
Ri think there is a complier error
Fvery likely
Fcuda 12.4 + accurate c++ tools you need
Fye it depends on each dataset
F400 not too much
Ras far as i can tell i have both
Ffor 10 images
Fbut inaccurate versions very likely
Rhow do i check
FYouTubeSECourses
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...
Fi am still using this video config
Ri did that. not sure what i missed
Ris there a way to check what i missed installing
Fsadly not
Fthese compiler stuff so so hard to solve
Fsometimes i am spending 10 hours to make scripts work
