ace-studio-v3-mashup / docker-compose.yml

Sync source

dd32171 unverified about 1 month ago

2.2 kB

# =============================================================================
# ACE-Step 1.5 — Docker Compose (generic CUDA)
# =============================================================================
#
# Usage:
#   docker compose up              # Start Gradio UI (default)
#   docker compose up --build      # Build and start
#   docker compose up -d           # Run in background
#   docker compose down            # Stop
#   docker compose logs -f         # View logs
#
#   # Start REST API server instead:
#   ACESTEP_MODE=api docker compose up
#
# =============================================================================

services:
  acestep:
    build:
      context: .
      dockerfile: Dockerfile
    image: ghcr.io/ace-step/ace-step-1.5:latest
    container_name: acestep

    # ---- GPU access ----
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    # ---- Environment ----
    env_file:
      - path: .env
        required: false
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - ACESTEP_MODE=${ACESTEP_MODE:-gradio}
      - ACESTEP_LLM_BACKEND=${ACESTEP_LLM_BACKEND:-pt}
      - ACESTEP_INIT_SERVICE=${ACESTEP_INIT_SERVICE:-true}
      - ACESTEP_CONFIG_PATH=${ACESTEP_CONFIG_PATH:-acestep-v15-turbo}
      - ACESTEP_LM_MODEL_PATH=${ACESTEP_LM_MODEL_PATH:-acestep-5Hz-lm-4B}
      - ACESTEP_EXTRA_ARGS=${ACESTEP_EXTRA_ARGS:-}
      - TOKENIZERS_PARALLELISM=false

    # ---- Ports ----
    ports:
      - "${GRADIO_PORT:-7860}:7860"
      - "${API_PORT:-8001}:8001"

    # ---- Persistent volumes ----
    volumes:
      - ./checkpoints:/app/checkpoints
      - hf_cache:/root/.cache/huggingface
      - ./gradio_outputs:/app/gradio_outputs
      - ./output:/app/output

    # ---- Shared memory for PyTorch DataLoader ----
    shm_size: "2gb"

    restart: unless-stopped

    healthcheck:
      test: >-
        curl -sf http://localhost:7860/ > /dev/null 2>&1 ||
        curl -sf http://localhost:8001/health > /dev/null 2>&1 ||
        exit 1
      interval: 30s
      timeout: 10s
      start_period: 300s
      retries: 3

volumes:
  hf_cache:

	# =============================================================================
	# ACE-Step 1.5 — Docker Compose (generic CUDA)
	# =============================================================================
	#
	# Usage:
	# docker compose up # Start Gradio UI (default)
	# docker compose up --build # Build and start
	# docker compose up -d # Run in background
	# docker compose down # Stop
	# docker compose logs -f # View logs
	#
	# # Start REST API server instead:
	# ACESTEP_MODE=api docker compose up
	#
	# =============================================================================

	services:
	acestep:
	build:
	context: .
	dockerfile: Dockerfile
	image: ghcr.io/ace-step/ace-step-1.5:latest
	container_name: acestep

	# ---- GPU access ----
	deploy:
	resources:
	reservations:
	devices:
	- driver: nvidia
	count: all
	capabilities: [gpu]

	# ---- Environment ----
	env_file:
	- path: .env
	required: false
	environment:
	- NVIDIA_VISIBLE_DEVICES=all
	- ACESTEP_MODE=${ACESTEP_MODE:-gradio}
	- ACESTEP_LLM_BACKEND=${ACESTEP_LLM_BACKEND:-pt}
	- ACESTEP_INIT_SERVICE=${ACESTEP_INIT_SERVICE:-true}
	- ACESTEP_CONFIG_PATH=${ACESTEP_CONFIG_PATH:-acestep-v15-turbo}
	- ACESTEP_LM_MODEL_PATH=${ACESTEP_LM_MODEL_PATH:-acestep-5Hz-lm-4B}
	- ACESTEP_EXTRA_ARGS=${ACESTEP_EXTRA_ARGS:-}
	- TOKENIZERS_PARALLELISM=false

	# ---- Ports ----
	ports:
	- "${GRADIO_PORT:-7860}:7860"
	- "${API_PORT:-8001}:8001"

	# ---- Persistent volumes ----
	volumes:
	- ./checkpoints:/app/checkpoints
	- hf_cache:/root/.cache/huggingface
	- ./gradio_outputs:/app/gradio_outputs
	- ./output:/app/output

	# ---- Shared memory for PyTorch DataLoader ----
	shm_size: "2gb"

	restart: unless-stopped

	healthcheck:
	test: >-
	curl -sf http://localhost:7860/ > /dev/null 2>&1 \|\|
	curl -sf http://localhost:8001/health > /dev/null 2>&1 \|\|
	exit 1
	interval: 30s
	timeout: 10s
	start_period: 300s
	retries: 3

	volumes:
	hf_cache: