Closed ProducerMatt closed 2 years ago
Hello.
On a first glance it seems that it is only required to override torch with cudaSupport=true. Here is draft PR - https://github.com/gbtb/nix-stable-diffusion/pull/2. I've made two variants, with and without mkl - nix develop .#invokeai-nvidia-mkl
and nix develop .#invokeai-nvidia
. Unfortunately, I can't confirm that this shells even build, because of some timeout errors when downloading magma (probably my ISP is messing it up). But you can try it nevertheless :smiley:
Wow, it worked first try! Thank you :)
Great! Btw, which one of the devshells worked for you - with or without mkl?
Both of them appear to work! But what does MKL change?
Both of them appear to work! But what does MKL change?
It changes blas implementation, it appears there is a bug in opencv4 that causes conflict of two incompatible blas versions. On my machine invokeai works only if I replace blas in opencv4 with different provider.
Thank you for the testing! I will clean up my PR a little bit and merge into master.
PR merged, slightly changed shell name - nix develop .#invokeai.nvidia
I have an Nvidia GPU, if you put up an Nvidia branch I'd happily test it for you. In the meantime I'll take a jab at converting it myself, but I've never worked with a flake like this before so idk how far I'll get on it.