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

[InTheFuture7]

Hello, this is a great job! However, when I was training with the vg100k dataset, I encountered an error: RuntimeError: CUDA error: no kernel image is available for execution on the device. After reviewing some information, I found that the possible reason is that the pytorch version does not support my gpu. Have you ever encountered this problem and is there any solution?

My gpu is GeForce RTX 4090. The torch version and other python extensions are consistent with requeirments.txt.

[InTheFuture7]

run torch.tensor([1.0, 2.0]).cuda() return RuntimeError: CUDA error: no kernel image is available for execution on the device

run torch.cuda.is_available() return True

[InTheFuture7]

I just ran command python collect_env.py, which can output relevant system environment infomation. I got this info :NVIDIA GeForce RTX 4090 with CUDA capability sm_89 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70. So, I will try to install a newer version of pytorch.

[InTheFuture7]

It works. The model is training!

I choose to install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1.