Getting PyTorch
Easiest: Use pre-built PyTorch
WEAVE provides Python environments with PyTorch and other LLNL-developed tools pre-installed:
Install a specific version
You will have some control over the PyTorch version if you install it yourself. How you install PyTorch depends on the system architecture of the machine you're on.
- AMD GPU systems (Tioga, Tuolumne, RZAdams, RZVernal, El Capitan, Tenaya):
- NVIDIA GPU systems (Matrix, RZVector)
- CPU-only systems (Dane, RZHound, etc.)
Full control: build PyTorch from source
Most users will not need this option. Resources coming soon for advanced users.
Running PyTorch
Consider using hpc-launcher from the LBANN project to launch PyTorch jobs. Resources coming soon.
General Python User Resources
Livermore Computing (LC) maintains Python and a set of site-specific packages (modules) on all production TOSS and CORAL systems. Please see our Python documentation page, which includes information on the following.
Interactive development
To use Jupyter notebooks on LC systems, use the LC Orbit service:
Python environment management
- System installations of Python and accessing them via Lmod
- Setting up a Python environment (with virtualenv or Conda) and working with site packages (and installing your own)
Additional resources
- Running python scripts, at scale, and in an HPC batch job
- Debugging python with TotalView and pdb
- Integrated Development Environments (IDEs)
- Spindle tool: Important for El Cap/Tuolumne users loading Python packages in large scale jobs