provider_id: my-mac-01
provider_token: <token minted on first admission>
coordinator: wss://coordinator.malibu.tech/ws/provider
model:
id: mlx-community/Qwen2.5-7B-Instruct-4bit
# Or a list to hold multiple warm:
# ids:
# - mlx-community/Qwen2.5-7B-Instruct-4bit
# - mlx-community/Llama-3.2-3B-Instruct-4bit
runtime:
max_context: 8192 # tokens
max_batch: 1
gpu_memory_reserve_gb: 2 # leave this much unified memory for the OS
heartbeat:
interval_s: 30
autoupdate:
enabled: true # opt out only for pinned/audited fleets
autotune:
enabled: true # run recommend on first launch
min_tps: 15 # skip candidates below this
max_ttft_ms: 4000 # skip candidates above this on 4K prompts
logging:
level: info
file: ~/Library/Logs/malibu/malibu-cli.log