- name: Setup Qwen2.5 - Coder (FIM - LLAMA.cpp)
  hosts: qwen2_host
  vars:
    podman_user: "{{ qwen2_user | default('qwen2') }}"
    podman_group: "{{ qwen2_group | default('qwen2') }}"
    podman_extra_groups: "users,video"
    podman_user_home: "{{ qwen2_home | default('/home/qwen2') }}"
    llama_model: "{{ qwen2_model | default('ggml-org/Qwen2.5-Coder-1.5B-Q8_0-GGUF:Q8_0') }}"
    podman_user_folders:
      - models
      - .cache
      - .cache/llama.cpp
    pods:
      - name: qwen2
        repo:
          image: docker.io/rocm/llama.cpp:llama.cpp-b6652.amd0_rocm7.0.0_ubuntu24.04_server
        env:
          LLAMA_ARG_THREADS: 6
          LLAMA_ARG_PARALLEL: 1
          LLAMA_ARG_CACHE: "true"
          LLAMA_ARG_CTX_SIZE: 8192
        ports:
          - "0.0.0.0:{{ qwen2_port | default(8012) }}:8012/tcp"
        command:
          - "-hf"
          - "{{ llama_model }}"
          - "--fim-qwen-1.5b-default"
        volumes:
          - "{{ podman_user_home }}/models:/models:Z"
          - "{{ podman_user_home }}/.cache:/app/.cache:Z"
        device:
          - "/dev/kfd:/dev/kfd:rw"
          - "/dev/dri:/dev/dri:rw"
  tasks:
    - name: Setup Pods
      ansible.builtin.include_tasks:
        file: tasks/podman.yml