install.sh — SDK Installer

The install.sh script inspects your system and installs everything needed to run AI models on Axelera hardware. You run it once after cloning the SDK, and again each time you switch to a different SDK release.

Quick reference

# Full install with sample videos
./install.sh --all --media

# Full install without sample videos
./install.sh --all

# Runtime only (no development tools)
./install.sh --runtime --no-development

# See all options
./install.sh --help

What it installs

The installer sets up three components:

Component	Location	Purpose
PCIe driver	`/lib/modules/`	Kernel module for Metis hardware communication
Runtime libraries	`/opt/axelera/\<version\>/`	Host-side libraries for device access and model execution
Python virtual environment	`~/.cache/axelera/venvs/` (symlinked to `./venv`)	Python packages for the complete SDK toolchain

PCIe driver

Installed to the system location. Contains minimal functionality that rarely changes between SDK releases. Requires sudo privileges.

Runtime libraries

Installed to /opt/axelera/ in version-specific directories. This means you can have multiple SDK versions installed side by side — each with its own runtime libraries.

Python virtual environment

Installed to ~/.cache/axelera/venvs/ and symlinked to ./venv in the SDK directory. Contains all Python dependencies for compilation, inference, and deployment.

Always activate the environment

After installation, you must activate the virtual environment before using any SDK tool:

source venv/bin/activate

Without this, commands like inference.py and axdevice will not work.

Common options

Option	Purpose
`--all`	Install all components (driver, runtime, Python environment)
`--media`	Download sample videos for testing (traffic, pedestrians, etc.)
`--runtime --no-development`	Install only what's needed to run models (skip compiler/development tools)
`--help`	Show all available options

When to re-run

You must re-run install.sh when:

Switching SDK versions — Each release may have different dependencies
After a fresh clone — The virtual environment doesn't exist yet
After system updates — Kernel updates may require driver reinstallation

You do not need to re-run when:

Running inference with an already-installed SDK version
Changing models or pipelines
Updating your own application code

Repository structure

After cloning, the SDK repository contains:

Path	What it is
`install.sh`	This installer
`cfg/`	Configuration files per platform (e.g., `config-ubuntu-2204-amd64.yaml`)
`inference.py`	Pipeline evaluation tool
`deploy.py`	Pipeline deployment tool
`ax_models/`	Model Zoo — pre-trained models and pipelines
`examples/`	Example applications
`docs/`	Documentation

Multiple SDK versions

You can install multiple SDK versions on the same system:

# Clone and install v1.5
git clone https://github.com/axelera-ai-hub/voyager-sdk.git sdk-v1.5
cd sdk-v1.5
./install.sh --all

# Clone and install v1.4
cd ..
git clone https://github.com/axelera-ai-hub/voyager-sdk.git sdk-v1.4
cd sdk-v1.4
git checkout release/v1.4.0
./install.sh --all

Each version has its own venv/ and runtime libraries. Switch between them by activating the appropriate environment.

important

When switching SDK versions, always deactivate the current environment first, then source venv/bin/activate in the new SDK directory.

Quick reference​

What it installs​

PCIe driver​

Runtime libraries​

Python virtual environment​

Common options​

When to re-run​

Repository structure​

Multiple SDK versions​

See also​