訓練時遇到了 RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)錯誤

[chenluchi420]

遇到了和 #4 一樣的問題，修正成"with autocast()"之後出現了下面錯誤：

本問題，目前無法排除

[raviakash]

I also get the same error when running on RTX 4070 GPU. This is an issue with PyTorch and nvrtc for the new RTX 40 series GPUs. Please check. RTX 40 series GPUs support sm89 architecture that's explicitly supported only by cuda 11.8. On Detectron 2 installation pages it says The support is only until cuda 11.3 Check. I ran the same thing on the RTX 30 series GPU and the model ran completely fine without any errors.