NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.

[wvffle]

After running the service I get:

Oct 04 13:19:39 llm sd-webui-forge-start[394]: /nix/store/a1jmpf3hja9g4rx7lwxza4gwng9vscj5-python3-3.11.13-env/lib/python3.11/site-packages/torch/cuda/__init__.py:209: UserWarning:
Oct 04 13:19:39 llm sd-webui-forge-start[394]: NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
Oct 04 13:19:39 llm sd-webui-forge-start[394]: The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
Oct 04 13:19:39 llm sd-webui-forge-start[394]: If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

When trying to generate an image i get the followirng error:

Oct 04 15:02:41 llm sd-webui-forge-start[394]: RuntimeError: CUDA error: no kernel image is available for execution on the device
Oct 04 15:02:41 llm sd-webui-forge-start[394]: CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
Oct 04 15:02:41 llm sd-webui-forge-start[394]: For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Oct 04 15:02:41 llm sd-webui-forge-start[394]: Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Could you please update torch?

[Janrupf]

Oh sorry, somehow overlooked this issue. The pytorch version is taken from an automated pip install (so version is selected by pip install automatically).

However, it can technically be overwritten - since I can't test this, would you be willing to PR the required changes?