holoscan-install-wheel

Install Holoscan SDK Python wheel via pip into a venv. Use for Python installs; not for native C++/apt or Conda installs.

2,779stars322forksUpdated 8/4/2026

Data Science & ML #installation #nvidia #pip #cuda #python

Security Assessment

Safe(90/100)

Security Score90/100

About holoscan-install-wheel

The holoscan-install-wheel skill guides an agent through installing the NVIDIA Holoscan SDK Python bindings via the pip wheel into a dedicated virtual environment. It solves the recurring friction of Holoscan Python setup, where the correct wheel depends on the host's CUDA driver, system-wide pip is blocked on newer Ubuntu, and stack-size limits can cause runtime warnings or segfaults. The skill scopes itself narrowly to Python installs and explicitly defers native C++/apt or Conda installs to sibling skills.

The workflow first consults the official NVIDIA installation docs (treating them as authoritative over the skill text), determines the CUDA variant from nvidia-smi to choose between the holoscan-cu12 and holoscan-cu13 packages, creates and activates a venv, installs the wheel, and verifies the install by importing the module and running the official hello_world and video_replayer examples pinned to the installed version tag. It documents prerequisites (Linux x86_64 with an NVIDIA GPU, CUDA Toolkit on PATH, Python 3.10-3.13, network access to PyPI), the ulimit -s 32768 recommendation, and how to source example data. Notably, it fetches remote example scripts over HTTPS and instructs the agent to warn the user before downloading and executing them.

Target users are developers and engineers building Python-based streaming analytics and medical-device applications on Holoscan who need a reliable, reproducible pip-wheel install path. It is most useful in fresh environments, CI provisioning, or when onboarding to the SDK for the first time.

FAQ

How do I know whether to install holoscan-cu12 or holoscan-cu13?

Read the CUDA Version field from nvidia-smi. A 13.x+ driver uses holoscan-cu13; any 12.x driver uses holoscan-cu12. The two packages are mutually exclusive and must match the host CUDA major.

What are the prerequisites?

Linux x86_64 with an NVIDIA GPU and driver, the CUDA Toolkit on PATH matching the host CUDA major, Python 3.10-3.13 with venv available, and network access to PyPI and docs.nvidia.com.

Does the wheel install everything Holoscan offers?

No. It is Python-only. For C++ headers and libraries you must pair it with the holoscan-install-debian skill. The video_replayer sample data also ships only with the Debian package, so you must set HOLOSCAN_INPUT_PATH otherwise.

Why does the skill recommend ulimit -s 32768?

Without a raised stack size some Holoscan apps emit a stack-size warning or, in rarer cases, segfault. Setting it in every shell that runs Holoscan suppresses the warning.

Does it run remote code?

It downloads official NVIDIA example scripts over HTTPS pinned to the installed version tag and runs them to verify the install. The skill instructs the agent to tell the user before downloading and executing them, and to skip that step if the user declines.