[
  {
    "label": "medium-slot1-priority3-draft-cuda0",
    "model": "pure",
    "ctx": 160000,
    "single": false,
    "split": "70,30",
    "ubatch": 128,
    "parallel": 1,
    "draft_max": 7,
    "draft_kv": "q4_0",
    "draft_device": "CUDA0",
    "quality": false,
    "prompts": [
      4096
    ],
    "minimum_headroom_mib": 512
  }
]