[Unit]
Description=Local Qwen3.5-2B chat server
StartLimitIntervalSec=300
StartLimitBurst=3

[Service]
Type=simple
WorkingDirectory=%h/local-llm
Environment=OMP_NUM_THREADS=2
Environment=OPENBLAS_NUM_THREADS=2
Environment=HF_HUB_OFFLINE=1
Environment=HF_HUB_DISABLE_TELEMETRY=1
ExecStart=%h/local-llm/runtime/build/bin/llama-server \
    --model %h/local-llm/models/Qwen3.5-2B-Q4_K_M.gguf \
    --mmproj %h/local-llm/models/mmproj-F16.gguf \
    --no-mmproj-offload --image-min-tokens 64 --image-max-tokens 512 \
    --alias Qwen3.5-2B \
    --host 127.0.0.1 --port 8092 --cors-origins localhost \
    --ctx-size 4096 --parallel 1 \
    --threads 2 --threads-batch 2 \
    --batch-size 128 --ubatch-size 64 \
    --cache-ram 256 --n-predict 512 \
    --jinja --reasoning off \
    --temp 0.7 --top-p 0.8 --top-k 20 --min-p 0 \
    --presence-penalty 1.5 \
    --offline --no-ui-mcp-proxy \
    --path %h/local-llm/runtime/webui \
    --ui-config-file %h/local-llm/config/ui.json
Restart=on-failure
RestartSec=10
TimeoutStopSec=30
KillSignal=SIGINT
Nice=10
MemoryHigh=2300M
MemoryMax=3G
MemorySwapMax=0
NoNewPrivileges=true
UMask=0077

[Install]
WantedBy=default.target