I'm trying to run setup and I'm having trouble finding the torch header file, my torch version is 0.4.0, and I've seen some people say that using git checkout ac045b6072b9bc3454fb9f9f9f17674f0d59373789 to switch to a branch that supports 0.4.0 solves the problem.
But the problem is that there are commands like __shfl_down in that branch, which gives an error when making, and I found out that this command is deprecated in CUDA 9.0+. Since my hardware is a GTX 3080ti, which is only supported by CUDA 11.x+, I can't use the above branch either. Anyway, I can't find a compatible version of CUDA11.x+pytorch0.4.0, what should I do?
我在尝试运行setup遇到了找不到torch头文件的问题,我的torch版本为0.4.0,我看到之前有人说使用
git checkout ac045b6072b9bc3454fb9f9f17674f0d59373789
切换到支持0.4.0的分支就可以解决问题 但问题是那个分支中有使用__shfl_down这样的命令,在make时会报错,我查到这个命令在CUDA9.0以上被弃用了。由于我的硬件是GTX 3080ti,只有CUDA11.x以上的版本才能支持,所以我也不能采用以上的分支。总之,我找不到兼容CUDA11.x+pytorch0.4.0的版本,我该怎么办?I'm trying to run setup and I'm having trouble finding the torch header file, my torch version is 0.4.0, and I've seen some people say that using
git checkout ac045b6072b9bc3454fb9f9f9f17674f0d59373789
to switch to a branch that supports 0.4.0 solves the problem. But the problem is that there are commands like __shfl_down in that branch, which gives an error when making, and I found out that this command is deprecated in CUDA 9.0+. Since my hardware is a GTX 3080ti, which is only supported by CUDA 11.x+, I can't use the above branch either. Anyway, I can't find a compatible version of CUDA11.x+pytorch0.4.0, what should I do?