Install the Voyager SDK
These instructions are for Linux (Ubuntu 22.04+). For Windows, see the Windows Setup Guide — you will need WSL2 configured first.
Quick start
For quick starting your Voyager SDK installation, run:
git clone https://github.com/axelera-ai-hub/voyager-sdk.git
cd voyager-sdk
./quick-start.sh
The script will install all required dependencies and create a python virtual environment with the full SDK dependencies.
It installs the SDK version recorded in VERSION.TXT.
If you want more control, continue reading.
In-depth guide
The install.sh SDK installer has been removed. Use ./quick-start.sh (above) or the pip instructions below instead.
The supported installation path is Python pip.
| Path | Best for | What it installs |
|---|---|---|
| Python pip | New installations, SDK 1.6+ | axelera-rt, axelera-devkit via pip |
Prerequisites
- Ubuntu 22.04 or later
- Python 3.10, 3.11, 3.12, or 3.13
- Git installed
- Internet connection
- sudo privileges
- Metis hardware installed and powered on (see Hardware Installation)
Python pip installation
Always refer to the Compatibility Matrix for the versions
Step 1: Clone the repository and install dependencies
git clone https://github.com/axelera-ai-hub/voyager-sdk.git
cd voyager-sdk
Install system dependencies using the provided script:
./install-dependencies.sh
Step 2: Create a virtual environment and install
python3 -m venv axelera-env
source axelera-env/bin/activate
pip install --extra-index-url https://software.axelera.ai/artifactory/api/pypi/axelera-pypi/simple axelera-rt axelera-devkit[all]
make operators
Use a dedicated virtual environment for each SDK version or project to avoid dependency conflicts.
Step 3: Install the Metis kernel driver
If installation of the driver fails, ensure your kernel headers are present: sudo apt-get install -y linux-headers-$(uname -r)
There are two ways of installing the driver, Option a) is the easiest.
Option a) use axdevice
axdevice driver --install
By default, axdevice will default to installing the recommended driver version
Option b) use apt
# Add the Axelera apt repository
sudo sh -c "curl -fsSL https://software.axelera.ai/artifactory/api/security/keypair/axelera/public | gpg --dearmor -o /etc/apt/keyrings/axelera.gpg"
# Ubuntu 22.04
sudo sh -c "echo 'deb [signed-by=/etc/apt/keyrings/axelera.gpg] https://software.axelera.ai/artifactory/axelera-apt-source ubuntu22 main' > /etc/apt/sources.list.d/axelera.list"
# Ubuntu 24.04
sudo sh -c "echo 'deb [signed-by=/etc/apt/keyrings/axelera.gpg] https://software.axelera.ai/artifactory/axelera-apt-source ubuntu24 main' > /etc/apt/sources.list.d/axelera.list"
sudo apt-get update
sudo apt-get install -y metis-dkms
Verify the driver is loaded:
lsmod | grep metis
Verify the loaded driver version:
modinfo metis | grep ^version:
Reload the driver:
sudo modprobe -r metis
sudo modprobe metis
Step 4: Verify installation
axdevice
Expected output lists detected Metis devices with their firmware versions. If no devices are listed, run axdevice --refresh and check that the driver is loaded.
Slimmed-down environments
For deployment-only or compile-only machines:
| Environment | Install command |
|---|---|
| Runtime only - axrunmodel | pip install --extra-index-url https://software.axelera.ai/artifactory/api/pypi/axelera-pypi/simple axelera-rt |
| Runtime only - inference.py | pip install --extra-index-url https://software.axelera.ai/artifactory/api/pypi/axelera-pypi/simple axelera-rt[all] |
| Compiler only - deploy.py | pip install --extra-index-url https://software.axelera.ai/artifactory/api/pypi/axelera-pypi/simple axelera-devkit[all] |
The [all] extra is what makes deploy.py and inference.py work. Plain
axelera-rt and axelera-devkit carry the runtime and the compiler, but not
the Python libraries axelera.app imports on top of them, such as
strictyaml and opencv-python. Install without the extra only if you are
calling the compiler or runtime APIs directly.
Proceed to Verify Your Setup.
SDK Installer (removed)
install.sh has been removed (along with launch-docker.sh, installer_support.py and the cfg/ files it used). Use ./quick-start.sh or the Python pip installation above.
- Virtual environments are now created by
quick-start.shas./venv_\<version\>; activate withsource venv_\<version\>/bin/activate. - To switch SDK versions,
git checkoutthe desired tag and run./quick-start.shagain. - Sample videos (formerly
--media) are no longer downloaded by an installer; useaxdownloadmediato list and download them.
Next steps
Your SDK is installed and activated. Proceed to Verify Your Setup.