Add tracked systemd unit template for llama-server — previously only existed as commands run live on BigBoy

ansible/templates/llama-server.service.j2

@@ -0,0 +1,54 @@
+# llama-server.service — systemd unit for BigBoy's llama.cpp inference server
+#
+# This is a Jinja2 template (Ansible's `template` module convention),
+# ready to use once Phase 5 is formalized into an actual Ansible role.
+# Until then, it's the tracked, reproducible reference for what's
+# actually configured live on BigBoy — previously this only existed as
+# commands typed directly during the deployment session, not committed
+# anywhere.
+#
+# Variables reference group_vars/bigboy.yml where they already exist
+# (llama_server_host, llama_server_port, model_repo, model_quant,
+# bigboy_admin_user). One deliberate exception: the API key is NOT
+# templated from group_vars — see the note below.
+#
+# Manual deployment (until this is wired into a real Ansible role):
+#   1. Fill in the Jinja2 variables below with their actual values
+#      (or render this file with `ansible-playbook --check` / a
+#      one-off `ansible.builtin.template` task pointed at this file)
+#   2. Copy the rendered result to /etc/systemd/system/llama-server.service
+#   3. sudo systemctl daemon-reload
+#   4. sudo systemctl enable --now llama-server
+
+[Unit]
+Description=llama.cpp server ({{ model_repo }}, {{ model_quant }})
+After=network.target
+
+[Service]
+Type=simple
+User={{ bigboy_admin_user }}
+
+# API key: deliberately NOT a group_vars variable. This template is
+# committed to a public repo — hardcoding even a placeholder that might
+# later get filled with a real key and accidentally committed is a real
+# leak risk. Supply this via ansible-vault, an environment file loaded
+# with EnvironmentFile= below, or a gitignored file following the same
+# pattern as ansible/keys/workbench.pub — never a plain variable here.
+#
+# Example using an EnvironmentFile (uncomment and adjust once the key
+# file exists, e.g. /etc/llama-server/api-key.env containing
+# LLAMA_API_KEY=<the real key>, mode 0600, owned by root):
+# EnvironmentFile=/etc/llama-server/api-key.env
+
+ExecStart=/home/{{ bigboy_admin_user }}/llama.cpp/build/bin/llama-server \
+  -hf {{ model_repo }}:{{ model_quant }} \
+  -ngl 99 --jinja \
+  --host {{ llama_server_host }} --port {{ llama_server_port }} \
+  --cont-batching --parallel 4 \
+  --api-key ${LLAMA_API_KEY}
+
+Restart=on-failure
+RestartSec=5
+
+[Install]
+WantedBy=multi-user.target