[Unit]
Description=Foxhunt ML Services (TLOB, MAMBA, DQN, PPO)
After=foxhunt-core.service
Requires=foxhunt-core.service
Documentation=https://github.com/foxhunt/docs

[Service]
Type=exec
User=foxhunt
Group=foxhunt
WorkingDirectory=/opt/foxhunt
ExecStart=/opt/foxhunt/bin/foxhunt-ml
ExecReload=/bin/kill -HUP $MAINPID
ExecStop=/bin/kill -TERM $MAINPID
TimeoutStopSec=30
Restart=always
RestartSec=10

# GPU-enabled CPU cores with NUMA awareness
CPUAffinity=6-9
SchedulingPolicy=batch
SchedulingPriority=0

# High memory for ML models
MemoryAccounting=yes
MemoryMax=8G
LimitNOFILE=65536
LimitMEMLOCK=infinity

# GPU access
DeviceAllow=/dev/nvidia0 rwm
DeviceAllow=/dev/nvidia-uvm rwm
DeviceAllow=/dev/nvidia-uvm-tools rwm
DeviceAllow=/dev/nvidiactl rwm

# Security
NoNewPrivileges=true
ProtectSystem=strict
ProtectHome=true
ReadWritePaths=/opt/foxhunt/data /var/log/foxhunt /tmp /dev/shm

# Environment
Environment=RUST_LOG=info
Environment=FOXHUNT_CONFIG=/opt/foxhunt/config/production.toml
Environment=FOXHUNT_CORE_ENDPOINT=http://localhost:8080
Environment=CUDA_VISIBLE_DEVICES=0
Environment=FOXHUNT_ENV=production

[Install]
WantedBy=multi-user.target