# ODS — CPU-Only Overlay (Linux / no GPU)
# Uses llama.cpp server CPU build. No NVIDIA/AMD/Intel GPU required.
# For low-RAM systems, set LLAMA_SERVER_MEMORY_LIMIT=4G in .env
#
# Use with: docker compose -f docker-compose.base.yml -f docker-compose.cpu.yml up -d

services:
  llama-server:
    image: ${LLAMA_SERVER_IMAGE:-ghcr.io/ggml-org/llama.cpp:server-b8248}
    command:
      - --model
      - /models/${GGUF_FILE:-Qwen3.5-9B-Q4_K_M.gguf}
      - --host
      - 0.0.0.0
      - --port
      - "8080"
      - --n-gpu-layers
      - "0"
      - --ctx-size
      - "${CTX_SIZE:-8192}"
      - --batch-size
      - "${LLAMA_BATCH_SIZE:-512}"
      - --threads
      - "${LLAMA_THREADS:-4}"
      - --parallel
      - "${LLAMA_PARALLEL:-1}"
      - --metrics
    deploy:
      resources:
        limits:
          cpus: '${LLAMA_CPU_LIMIT:-8.0}'
          memory: ${LLAMA_SERVER_MEMORY_LIMIT:-6G}
        reservations:
          cpus: '${LLAMA_CPU_RESERVATION:-1.0}'
          memory: 2G