Install for other use cases (containers, HPC clusters, etc.)#

Prerequisites#

  • A system equipped with an NVIDIA GPU and Driver installed

  • CUDA versions: 11.3, 11.6, 11.7, 11.8, 12.1

  • Python: 3.7 - 3.11

Note

Create your own virtual environment using your preferred virtual environment manager. Ensure that the appropriate version of PyTorch is installed in your virtual environment.

Installing DeepView.Profile#

Run the following command to install DeepView.Profile

pip install deepview_profile

For users with PyTorch versions <2.2.0

Due to an issue with PyTorch profiler (which DeepView partially depends on to function), you would need to install an older version of DeepView for profiling to work. The issue relates to PyTorch’s handling of CUPTI contexts, so you will likely get CUPTI_ERROR_NOT_INITIALIZED or CUPTI_ERROR_MULTIPLE_SUBSCRIBERS_NOT_SUPPORTED if you have the incorrect combination of DeepView and PyTorch installed. This error is not present in the latest version of PyTorch.

  • If you have PyTorch version <2.0.1, install DeepView with the following:

    pip install deepview_profile==0.1.4
    
  • If you have PyTorch versions between 2.1.0 to 2.1.2, please either upgrade (to >=2.2.0) or downgrade (to <2.1.0) to a compatible version.

Installing the VSCode Extension#

Install DeepView.Explore (the VSCode extension) either from the Visual Studio Code Marketplace or install it directly from Visual Studio Code.

Run your first analysis#

Now you are ready for your first analysis!