docker-compose.gpu.yml

# GPU override for Ollama — combined with the base file via:
#
#   docker compose -f docker-compose.yml -f docker-compose.gpu.yml --profile gpu up -d
#
# Gotcha (verified 2026-05-20): the `deploy.resources.reservations.devices`
# block is silently no-op'd if `nvidia-container-runtime` is NOT the configured
# Docker runtime. Verify the host BEFORE invoking the gpu profile:
#
#   docker info | grep -i runtime    # must show "nvidia"
#   nvidia-smi                       # must show the GPU
#
# After bringing the stack up, confirm GPU passthrough inside the container:
#
#   docker exec forecastlab-ollama nvidia-smi
#
# See `docs/rag-ollama-setup.md` for embedding-model pull instructions
# (`ollama pull nomic-embed-text`) and the in-cluster `http://ollama:11434` URL.

services:
  ollama:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]