- name: Setup Qwen3.6 - Coder (FIM - LLAMA.cpp)
  hosts: qwen3_6_host
  vars:
    podman_user: "{{ qwen3_Z6_user | default('qwen36') }}"
    podman_group: "{{ qwen3_6_group | default('qwen36') }}"
    podman_extra_groups: "users,video"
    podman_user_home: "{{ qwen3_6_home | default('/home/qwen36') }}"
    llama_model: "{{ qwen3_6_model | default('bartowski/Qwen_Qwen3.6-35B-A3B-GGUF:IQ4_XS') }}"
    __qwen3_6_context_length__: "{{ qwen3_6_context_length | default(65537) }}"
    podman_user_folders:
      - models
      - .cache
      - .cache/llama.cpp
    pods:
      - name: qwen3.6
        repo:
          image: ghcr.io/ggml-org/llama.cpp:server-cuda
        ports:
          - "0.0.0.0:{{ qwen3_6_port | default(8012) }}:8080/tcp"
        command:
          - "-hf"
          - "{{ llama_model }}"
          - "-ngl"
          - "40"
          - "-ncmoe"
          - "32"
          - "-c"
          - "{{ __qwen3_6_context_length__ }}"
          - "-np"
          - "1"
          - "-ub"
          - "512"
          - "-b"
          - "1024"
          - "-fa"
          - "on"
          - "--no-mmap"
          - "-ctk"
          - "iq4_nl"
          - "-ctv"
          - "iq4_nl"
          - "-cram"
          - "{{ __qwen3_6_context_length__ }}"
        volumes:
          - "{{ podman_user_home }}/models:/models:Z"
          - "{{ podman_user_home }}/.cache:/app/.cache:Z"
        device:
          - "nvidia.com/gpu=all"
  tasks:
    - name: Setup Pods
      ansible.builtin.include_tasks:
        file: tasks/podman.yml