{"id":"Ww2cvMYs1XeMPv-lBg-WY","title":"docker-compose for patched hsdecide llama-server","language":"yaml","tags":[],"content":"services:\n  hsdecide:\n    # TODO: replace build context with the real fork once you have it\n    build: ./llama.cpp-hsdecide\n    image: local/llama.cpp-hsdecide\n    container_name: hsdecide\n    restart: unless-stopped\n    ports:\n      - \"8081:8080\"\n    volumes:\n      - ./models:/models\n    command: >\n      -m /models/hs-decide-1-flash-bf16.gguf\n      --host 0.0.0.0 --port 8080\n      --ctx-size 8192\n      -ngl 99\n    # ^ -ngl 99 offloads all layers to GPU; drop it for CPU-only\n    # this server exposes /v1/systemone (decision scoring), not chat completions\n","created_at":"2026-11-13T11:48:37.559Z"}