RFdiffusion did not install GPU version of Pytorch in the environment without CUDA

TL,DR. Will RFdiffusion only install Pytorch-CPU in the environment witout GPU and CUDA? I have solved my problems, and RFdiffusion works well in my study. However, I am very confused about this issue.

I have two environments to work on RFdiffusion.

First environment: In my local desktop computer, windows 10 system, as follows, I installed RFdiffussion by strictly following steps and everything works well. I got the Pytorch version is pytorch 1.9.1 py3.9_cuda11.1_cudnn8_0 pytorch .

Simply speaking, Pytorch-GPU version is installed in the environment with GPU.

$ nvidia-smi
Tue Mar 19 12:42:36 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 537.70                 Driver Version: 537.70       CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                     TCC/WDDM  | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  Quadro P4000                 WDDM  | 00000000:B3:00.0  On |                  N/A |
| 46%   35C    P0              28W / 105W |   1185MiB /  8192MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

Second environment: In the remote server, linux system, I have login accession who has CPU only and internet to use conda install. At this time, I also installed RFdiffusion by strictly following steps, but I find the pytorch is cpu, see pytorch 1.9.1 cpu_py39hc5866cc_3 defaults.

Simply speaking, only Pytorch-cpu is installed in the environment only with CPU, although I followed the total same installation steps.

In this remote server, linux system, I also have GPU accession who has GPU but no internet to use conda install. Therefore, when I run RFdiffusion in the remote GPU environment, RFdiffussion doesn't work due to Pytorch-CPU. Therefore, I have to uninstall the Pytorch-CPU and then reinstall Pytorch-GPU.

I feel confused, why Pytorch-CPU is installed in the environment without GPU, but Pytorch-GPU is installed in the environment with GPU, although the installation steps are totally same. Is it because the installation will detect environment by itself and decide to install CPU version or GPU version according to the environment by itself?

Thanks very much!

The payload names no repository files, tests, or entry points. Start by comparing the documented RFdiffusion installation steps and Conda's PyTorch package resolution in the two environments; done means establishing whether CPU/GPU selection is intentional and documenting any required offline GPU setup.