目前配置,cuda=11.1+与之匹配的cudnn+python3.7+pytorch1.8.0
在上面找到自己安装的torch版本,点击打开链接。我用的是torch-1.7.1+cu102(conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch),显示如下:torch_cluster-1.5.8-cp36-cp36m-linux_x86_64.whltorch_cluster-1.5.8-cp36-cp36m-win_amd64.whltorch_cluster-1.5.8-cp37-cp37m-linux_x86_64.whltorch_cluster-1.5.8-cp37-cp37m-win_amd64.whltorch_cluster-1.5.8-cp38-cp38-linux_x86_64.whltorch_cluster-1.5.8-cp38-cp38-win_amd64.whltorch_cluster-1.5.9-cp36-cp36m-linux_x86_64.whltorch_cluster-1.5.9-cp36-cp36m-win_amd64.whltorch_cluster-1.5.9-cp37-cp37m-linux_x86_64.whltorch_cluster-1.5.9-cp37-cp37m-win_amd64.whltorch_cluster-1.5.9-cp38-cp38-linux_x86_64.whltorch_cluster-1.5.9-cp38-cp38-win_amd64.whltorch_scatter-2.0.5-cp36-cp36m-linux_x86_64.whltorch_scatter-2.0.5-cp36-cp36m-win_amd64.whltorch_scatter-2.0.5-cp37-cp37m-linux_x86_64.whltorch_scatter-2.0.5-cp37-cp37m-win_amd64.whltorch_scatter-2.0.5-cp38-cp38-linux_x86_64.whltorch_scatter-2.0.5-cp38-cp38-win_amd64.whltorch_scatter-2.0.6-cp36-cp36m-linux_x86_64.whltorch_scatter-2.0.6-cp36-cp36m-win_amd64.whltorch_scatter-2.0.6-cp37-cp37m-linux_x86_64.whltorch_scatter-2.0.6-cp37-cp37m-win_amd64.whltorch_scatter-2.0.6-cp38-cp38-linux_x86_64.whltorch_scatter-2.0.6-cp38-cp38-win_amd64.whltorch_scatter-2.0.7-cp36-cp36m-linux_x86_64.whltorch_scatter-2.0.7-cp36-cp36m-win_amd64.whltorch_scatter-2.0.7-cp37-cp37m-linux_x86_64.whltorch_scatter-2.0.7-cp37-cp37m-win_amd64.whltorch_scatter-2.0.7-cp38-cp38-linux_x86_64.whltorch_scatter-2.0.7-cp38-cp38-win_amd64.whltorch_sparse-0.6.8-cp36-cp36m-linux_x86_64.whltorch_sparse-0.6.8-cp36-cp36m-win_amd64.whltorch_sparse-0.6.8-cp37-cp37m-linux_x86_64.whltorch_sparse-0.6.8-cp37-cp37m-win_amd64.whltorch_sparse-0.6.8-cp38-cp38-linux_x86_64.whltorch_sparse-0.6.8-cp38-cp38-win_amd64.whltorch_sparse-0.6.9-cp36-cp36m-linux_x86_64.whltorch_sparse-0.6.9-cp36-cp36m-win_amd64.whltorch_sparse-0.6.9-cp37-cp37m-linux_x86_64.whltorch_sparse-0.6.9-cp37-cp37m-win_amd64.whltorch_sparse-0.6.9-cp38-cp38-linux_x86_64.whltorch_sparse-0.6.9-cp38-cp38-win_amd64.whltorch_spline_conv-1.2.0-cp36-cp36m-linux_x86_64.whltorch_spline_conv-1.2.0-cp36-cp36m-win_amd64.whltorch_spline_conv-1.2.0-cp37-cp37m-linux_x86_64.whltorch_spline_conv-1.2.0-cp37-cp37m-win_amd64.whltorch_spline_conv-1.2.0-cp38-cp38-linux_x86_64.whltorch_spline_conv-1.2.0-cp38-cp38-win_amd64.whltorch_spline_conv-1.2.1-cp36-cp36m-linux_x86_64.whltorch_spline_conv-1.2.1-cp36-cp36m-win_amd64.whltorch_spline_conv-1.2.1-cp37-cp37m-linux_x86_64.whltorch_spline_conv-1.2.1-cp37-cp37m-win_amd64.whltorch_spline_conv-1.2.1-cp38-cp38-linux_x86_64.whltorch_spline_conv-1.2.1-cp38-cp38-win_amd64.whl
按照自己的python版本、操作系统确定一个车轮。我puhon是3.6,ubuntu22,64位,选择了torch_scatter-2.0.6-cp36-cp36m-linux_x86_64.whl下载后: pip install torch_scatter-2.0.6-cp36-cp36m-linux_x86_64.whl安装成功!No module named 'torch_sparse' 方法同上,我选择torch_sparse-0.6.8-cp36-cp36m-linux_x86_64.whl下载后: pip install torch_sparse-0.6.8-cp36-cp36m-linux_x86_64.whl