ONNX

[motorist828]

Why can't I choose ONNX or Vulkan?
I use vega 56 gpu

[ForserX]

Oh. This is the problem of having multiple versions of Python.

U need open (working_dir)\repo in cmd and call:

py -3.10 -m venv onnx.venv
onnx.venv\Scripts\activate
pip install -r "..\requirements_onnx.txt"
pip install ort-nightly-directml==1.15.0.dev20230408001 --extra-index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/

Or wait next release

[motorist828]

On the last step I get the error

[ForserX]

pip install ort-nightly-directml==1.15.0.dev20230408001 --extra-index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/

[ForserX]

Sorry, my bad

[motorist828]

hmm, ONNX is still not available
Is the line "pip install -r "..\requirements_onnx.txt" correct?

i used "pip install -r "requirements_onnx.txt" but not sure if it's correct

[ForserX]

Oh... U need use workpath\repo at cmd directory

[motorist828]

ok i did all the steps for the repo folder but nothing changed

[ForserX]

Ohh. U have CUDA. Sorry. It was an instruction for the AMD version

py -3.10 -m venv cuda.venv
cuda.venv\Scripts\activate
pip install -r "..\requirements_cuda.txt"

[motorist828]

I don't have CUDA, On my system radeon rx vega 56
I don't know why it is displayed in the program

[ForserX]

Oh... Vega don't have "AMD" in GPU name. Okay. I will release a corrected version in an hour or two. I'm sorry for the inconvenience.

[ForserX]

https://github.com/ForserX/StableDiffusionUI/releases/tag/3.1.2

[motorist828]

I uninstalled the previous version, installed the new one by selecting ONNX during installation
But I am getting this error when I run the generation

I downloaded the diffusion scripts folder from the repository
now the generation starts but i can see that it only uses the cpu

[ForserX]

@motorist828 Monday is a hard day. Fixed the archive.

[motorist828]

unfortunately it still doesn't work for my GPU

[ForserX]

As I see, the generation has passed and there is a picture on the background.

[motorist828]

Yes, but it didn't use my GPU, msi afterburner graph shows 0% load, video ram was also not used

[ForserX]

If he hadn't used it, then you would have generated 5-7 minutes, not 33 seconds.

It's better to just look in the usual Windows task manager.

[motorist828]

such time because I used 128x128 pixels knowing about this problem

[ForserX]

All right, call in Host wnd

py -install ort-nightly-directml==1.15.0.dev20230408001 --extra-index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/ --force

After click "Break!" in main form and gen img again

[motorist828]

Great, now it works

[ForserX]

I was glad to help! Good luck in further generations. If you have any more questions, you can use our server on Discord!