MLX backend (Apple Silicon)¶

Outcome: Serve L3–L5 via mlx_lm.server.

Steps¶

pip install mlx-lm
mlx_lm.server --model mlx-community/Llama-3.2-3B-Instruct-4bit --port 11440

mlx:
  enabled: true
  base_url: http://127.0.0.1:11440
  models:
    L3: mlx-community/Llama-3.2-3B-Instruct-4bit

Restart daari; daari doctor shows an mlx check.

Chat with X-Daari-Meta: true — executor should reflect MLX when selected.