35 lines
672 B
Django/Jinja
35 lines
672 B
Django/Jinja
[Quadlet]
|
|
DefaultDependencies=false
|
|
|
|
[Unit]
|
|
Description=llama.cpp Model Server
|
|
|
|
[Container]
|
|
Image=ghcr.io/ggml-org/llama.cpp:{{ version['containers']['llama'] }}
|
|
|
|
ContainerName=llama
|
|
HostName=llama
|
|
PublishPort={{ services['llama']['ports']['http'] }}:9931/tcp
|
|
|
|
# iGPU access for Vulkan
|
|
AddDevice=/dev/dri:/dev/dri
|
|
PodmanArgs=--group-add keep-groups
|
|
|
|
Volume=%h/containers/llama/models:/models:ro
|
|
Volume=%h/containers/llama/config:/config:ro
|
|
|
|
Environment="TZ={{ timezone }}"
|
|
|
|
Exec=--models-preset /config/models.ini \
|
|
--models-max 2 \
|
|
--host 0.0.0.0 \
|
|
--port 9931
|
|
|
|
[Service]
|
|
Restart=always
|
|
RestartSec=10s
|
|
TimeoutStopSec=300
|
|
|
|
[Install]
|
|
WantedBy=default.target
|