i was using your settings for sdxl lora training
i was using your settings for sdxl lora training
Fah yes they are decent but not very best
Fbest is SDXL DreamBooth
Joh i see
JFyour install selections
For better
Fdelete your accelerator settings
Fit shows at the end where it is saved
Fdelete this
Jill try that, thank you
J@Furkan Gözükara SECourses hmm i still seem to be getting the error
Fyou have 3 gpus?
Jyes
Fwell your setup wrong
Fi can professionally help
For you can limit to 1 gpu
Feasily
Fset CUDA_VISIBLE_DEVICES=0
Fadd this to .bat file
Jyea, i understand how to limit it to one gpu, but i would like to set it up for 3 gpu's
Fare your gpus same?
Jis that do-able?
Fwhat are they
Jtheyre all 4090's
Fye i think should work
Jwhat would i need to change to get that to work?
FDon't know. I have to connect your pc and figure out
Jim not sure that my workplace would be comfortable with that, we deal with a lot of proprietary information, is there any other way to go about it?
L@Josh It's the pytorch package I think, all the libraries in python are a hellish mess.
You can try to find a package that contains the distributed compilation code, you can set your configuration to single_gpu or you can move it into WSL.
I think going WSL is the best option
You can try to find a package that contains the distributed compilation code, you can set your configuration to single_gpu or you can move it into WSL.
I think going WSL is the best option
JLYou use Win11 ?
Lyou can google for pytorch, windows, nccl .. maybe you find a working package or you can compile it.
Or on Windows 11 you install WSL-2 with Ubuntu or Debian and then use some linux/WSL tutorial. It should be easier to train that way and faster - nvidia cuda drivers work better in the linux environment
Or on Windows 11 you install WSL-2 with Ubuntu or Debian and then use some linux/WSL tutorial. It should be easier to train that way and faster - nvidia cuda drivers work better in the linux environment
Fi dont think so necessary
Fkohya already uses accelerator
Factually i should test this on my pc
Fi have rtx 3060 and 3090
LI asked about Win11 because it has a significantly better WSL support with native linux kernels, makes the choice easier
LNot for long, we're working on AI-supported OpenBSD drivers for all famous NVidia etc. drivers
LAfter that, why would one want to run Linux at all 
LEverybody: OpenBSD what's that?
Choose OpenBSD for your Unix needs. OpenBSD -- the world's simplest and most secure Unix-like OS. A safe alternatve to the frequent vulnerabilities and overengineering of Linux and related software like NGiNX & Apache (https://openbsd.org/papers/httpd-asiabsdcon2015.pdf), OpenSSL, iptables/nftables, systemd, BIND, Postfix, Docker and so on.
Choose OpenBSD for your Unix needs. OpenBSD -- the world's simplest and most secure Unix-like OS. A safe alternatve to the frequent vulnerabilities and overengineering of Linux and related software like NGiNX & Apache (https://openbsd.org/papers/httpd-asiabsdcon2015.pdf), OpenSSL, iptables/nftables, systemd, BIND, Postfix, Docker and so on.
LAlso yeah! Since this is
#aiLHas anybody tried and can vouch for https://github.com/BoxcarsAI/boxcars? I guess it's kinda like LangChain but the next step in the evolution (it's written in Ruby too so)
GitHub
Building applications with composability using Boxcars with LLM's. Inspired by LangChain. - GitHub - BoxcarsAI/boxcars: Building applications with composability using Boxcars with LLM&#...
DHey there, can someone help with my TensorRT exporter please? I want to use it for my SDXL Dreambooth model, but I just can't generate an engine for it, it fails with the following error:
I've searched around and people are saying that I should be on the 'dev' branch, which I am. Also I tried removing the --medvram
I installed a fresh installation of the A1111 with the .bat file from the Patreon, the same with TensorRT.
I'm using a 3060Ti 8GB VRAM
ERROR:root:Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)I've searched around and people are saying that I should be on the 'dev' branch, which I am. Also I tried removing the --medvram
I installed a fresh installation of the A1111 with the .bat file from the Patreon, the same with TensorRT.
I'm using a 3060Ti 8GB VRAM
Fupdate to the latest version automatic1111
Fremove --medvram and --lowvram
Fincrease your virtual RAM if you dont have sufficient RAM
DI think when I removed --medvram it was the same error and not anything about RAM. Let me check again
Fthose optimizations make it load
Fsome part gpu