docker-compose.gpu.yml

# GPU override for the Ollama runtime (NVIDIA).
#
# Requires on the HOST:
#   - an NVIDIA GPU + a recent driver
#   - the NVIDIA Container Toolkit configured for Docker:
#     https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
#
# Usage (Compose merges this file on top of the base one):
#   docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d --build
#
# Verify the GPU is visible inside the runtime:
#   docker compose exec ollama nvidia-smi
#
# CPU-only hosts should simply omit this file — the base stack runs as-is.

services:
  ollama:
    # Reserve the host GPU(s) for the Ollama container. Ollama auto-detects CUDA
    # and offloads the models onto the GPU; no application code changes needed.
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
      # Optional: force a max number of GPU layers (unset = as many as fit in VRAM).
      # OLLAMA_NUM_GPU: "999"