xudayemath/zflows repository overview

⁠zflows on CUDA 13

Docker image with zflows⁠ — self-contained PyTorch normalizing flows for unconditional energy-based sampling and Boltzmann generators — preinstalled on top of nvidia/cuda:13.2.1-devel-ubuntu24.04.

⁠Contents

• Ubuntu 24.04 + CUDA 13.2 (devel)
• Python 3.12 in a venv at /opt/torch (active by default)
• PyTorch (CUDA 13 / cu130 wheels) + torchvision
• numpy, scipy, scikit-learn, pandas, matplotlib, h5py, tqdm
• zflows (latest from PyPI)

⁠Usage

Interactive shell (GPU attached):

docker run --rm -it --gpus all xudayemath/zflows:cu13

Mount your project and work in it:

docker run --rm -it --gpus all -v "$PWD:/workspace" xudayemath/zflows:cu13

Python REPL:

docker run --rm -it --gpus all xudayemath/zflows:cu13 python

Check torch.compile works on the GPU:

docker run --rm -it --gpus all xudayemath/zflows:cu13 \
  python -c "import zflows; zflows.utils.check_compile_available()"

--gpus all requires the NVIDIA Container Toolkit on the host.

⁠Tags

• cu13, latest — CUDA 13.2, Python 3.12