Open wolfyangfeng opened 1 year ago
I am currently running the benchmarks on my 5700XT.
I am running stable-diffusion using pytorch 1.13.1+rocm5.2, and not using the directml fork. I am also running on Linux and not windows.
This doesnt work for me on windows directml either
working fine on directml with a 5700 XT
the command lines I use:
set COMMANDLINE_ARGS= --opt-split-attention --precision autocast --no-half-vae --opt-channelslast --no-download-sd-model --opt-sdp-no-mem-attention --lowvram --upcast-sampling --skip-torch-cuda-test --disable-nan-check --enable-insecure-extension-access --autolaunch --api --cors-allow-origins http://localhost:5173
(some commands may be redundant, but if it works, it works)
when I run this extension,it just trackback error"AttributeError: module 'torch._C' has no attribute '_cuda_resetPeakMemoryStats' ".
I ran the webui in windows10,with directml.