Skip to main content

Linux Installation & Setup

GoingMerry supports native bare-metal execution on Linux with hardware acceleration capabilities for CPU vectorization paths (AVX-512) and GPU interfaces (Nvidia CUDA and AMD ROCm).

1. Quick Install Script​

Once staging is complete, a one-line script will install the binary:

curl -fsSL https://dopove.com/install.sh | sh

2. Manual Installation​

If you prefer to download and install the standalone compiled binary directly:

Step A: Download the Binary​

Retrieve the compiled target from the registry distribution server:

sudo curl -L https://dopove.com/download/merry-linux-amd64 -o /usr/local/bin/merry
sudo chmod +x /usr/local/bin/merry

Step B: Configure systemd Service​

To ensure the runtime daemon operates as a persistent system background service, create a systemd configuration file at /etc/systemd/system/merry.service:

[Unit]
Description=GoingMerry Service
After=network-online.target

[Service]
ExecStart=/usr/local/bin/merry serve
User=merry
Group=merry
Restart=always
RestartSec=3
Environment="MERRY_HOST=127.0.0.1:11434"

[Install]
WantedBy=default.target

Ensure a dedicated system user exists:

sudo useradd -r -s /bin/false -U -m -d /usr/share/merry merry

Reload, enable, and start the background service:

sudo systemctl daemon-reload
sudo systemctl enable merry
sudo systemctl start merry

3. GPU Hardware Drivers Configuration​

GoingMerry will auto-detect compatible GPU hardware. If multiple adapters are active, layers are sharded dynamically.

Nvidia GPUs (CUDA)​

Verify that your NVIDIA CUDA toolkit is operational:

nvidia-smi

If your CUDA library paths are not loaded under standard system search paths, define the library path variables in the systemd service configurations:

Environment="LD_LIBRARY_PATH=/usr/local/cuda/lib64"

AMD GPUs (ROCm)​

For AMD graphics architectures, verify ROCm libraries are active:

rocminfo

Ensure the systemd user is added to the render group to allow GPU access:

sudo usermod -a -G render merry