RuntimeError: CUDA error: no kernel image is available for execution on the device

[MingyuYang-Yale]

Hi,

I followed the instructions on the GitHub README page to install xFuse. However, when I ran the command:

xfuse run my-config.toml --save-path my-run

I encountered the following error:

To resolve this, I attempted to upgrade Torch by running:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

After the upgrade, I encountered a different error:

Additionally, when I ran nvidia-smi, here is the GPU information:

Could you please provide guidance on how to resolve this issue?

Thank you!

Best,
Mingyu

[younng367]

I have solve it. First, you should check the whole erro issue. You might be met the version issue of pytorch. Because previous pytorch vesion not support your GPU, so you need to find the best version for your GPU. It usually pop up in you erro like"NVIDIA RTX A4500 with CUDA capability sm_86 is not compatible with the current PyTorch installation."(It is my erro, and my pytorch version is 2.3.1. I find it is a higher version for my GPU, so i downgraded it . Moverover, you need to download the corresponding version for torchaudio and torchvision).

[younng367]

My mail 839116394@qq.com, if i can help.