macros:
  llama: |
    llama-server --port ${PORT}

  # - small `n` are not recommended
  # - MoEs require long drafts
  # - dense models: can reduce `--draft-min` and `--draft-max`
  next_edit: |
    --spec-type ngram-mod
    --spec-ngram-size-n 24
    --draft-min 48
    --draft-max 64

models:
  sweep-next-edit-1.5b:
    name: Sweep Next-Edit 1.5B
    unlisted: true
    cmd: |
      ${llama} --hf-repo sweepai/sweep-next-edit-1.5B
      ${next_edit}
  sweep-next-edit-v2-7b:
    name: Sweep Next-Edit v2 7B
    unlisted: true
    cmd: |
      ${llama} --hf-repo henrik3/sweep-next-edit-v2-7B-GGUF
      ${next_edit}
  zeta2:
    name: Zeta 2
    unlisted: true
    cmd: |
      ${llama} --hf-repo bartowski/zed-industries_zeta-2-GGUF
      ${next_edit}

groups:
  edit_prediction:
    persistent: true # not unloaded by other groups
    swap: true # only run one model from this group
    exclusive: false
    members:
      - sweep-next-edit-1.5b
      - sweep-next-edit-v2-7b
      - zeta2

hooks:
  on_startup:
    preload:
      - sweep-next-edit-v2-7b