version: '3.7'
services:
  custom_whisper:
    container_name: custom_whisper
    image: ghcr.io/cociweb/amd64_cuda-standalone_whisper:latest
    ports:
      - 10300:10300
    volumes:
      - custom_whisper_data:/data
    restart: always
    command: --model custom --language hu --device cuda --beam-size 1 --compute-type float32 --custom_model_name "My customized Whisper model" --custom_model_url "https://huggingface.co/Hungarians/whisper-medium-cv16-hu/resolve/main/fp32/"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
# If you would like to limit your cpu/memory resources, uncomment (and adjust) the following lines:
#        limits:
#          cpus: "4.0"
#          memory: 8096M

volumes:
  custom_whisper_data: