[Unit]
Description=EgoData vLLM inference (Qwen3-VL-8B-FP8, port __VLLM_PORT__)
After=network.target

[Service]
Type=simple
WorkingDirectory=/tmp
ExecStart=/bin/bash "__PROJECT_ROOT__/scripts/run_vllm.sh"
Environment=EGODATA_VLLM_PORT=__VLLM_PORT__
Environment=EGODATA_VLLM_URL=__VLLM_URL__
Restart=on-failure
RestartSec=10
# 模型加载较慢,给足启动时间
TimeoutStartSec=600

[Install]
WantedBy=__WANTED_BY__