# llama.cpp quadlet container — the homelab local LLM inference server
[Unit]
Description=llama.cpp server for local inference (qwen3-8b)
Requires=llamacpp-network.net
After=network-online.target

[Container]
Image=docker.io/ggml-org/llama-cpp:0.1.43
ContainerName=llamacpp
Network=llamacpp-network
PublishPort=8081:8080
Volume=/srv/llama/models:/models:ro
Environment=CONTEXT_LENGTH=32768
Environment=BATCH_SIZE=512
Restart=always

[Service]
TimeoutStartSec=300

[Install]
WantedBy=default.target
# RESE-EDIT-SUMMARY-SENTINEL-b41d — phase 57 fixture marker: last line, outside the 24-token digest.