What problem does it solve?
Installing the Holoscan SDK Python bindings requires matching the correct pip wheel (holoscan-cu12 or holoscan-cu13) to the host CUDA version, setting up a virtual environment, and verifying the install — a process with several failure points such as CUDA mismatches, externally-managed-environment errors, and stack-size segfaults.
Core Features & Use Cases
- CUDA-aware wheel selection: Detects the host CUDA version via nvidia-smi and picks the matching holoscan-cu12 or holoscan-cu13 package.
- Guided venv setup and install: Creates and activates a Python virtual environment at ~/holoscan/venv and installs the wheel via pip, avoiding Ubuntu 24.04 system-pip restrictions.
- Verification and troubleshooting: Validates the install with a version check plus hello_world and headless video_replayer examples, and provides fixes for common errors like ImportError, stack-size warnings, and missing racerx data.
- Use Case: A developer building a medical imaging pipeline on an RTX 4090 workstation with CUDA 12.4 asks the agent to set up the Holoscan Python SDK; the agent installs holoscan-cu12 into a venv and confirms it works by running the examples.
Quick Start
Ask your agent to install the Holoscan SDK Python bindings with pip into a virtual environment and verify the installation works.