services: # StemEngine backend (GPU-enabled) stemengine: container_name: stemengine image: stemengine:latest pull_policy: never runtime: nvidia entrypoint: [] restart: unless-stopped networks: - t2_proxy security_opt: - no-new-privileges:true ports: - "8000:8000" labels: - traefik.enable=true - traefik.docker.network=t2_proxy - traefik.http.routers.stemengine-api.rule=Host(`${STEMENGINE_API_HOST:-stemengine-api.ts.net}`) - traefik.http.routers.stemengine-api.entrypoints=${TRAEFIK_ENTRYPOINTS:-websecure} - traefik.http.routers.stemengine-api.tls=true - traefik.http.routers.stemengine-api.service=stemengine-api-svc - traefik.http.services.stemengine-api-svc.loadbalancer.server.port=8000 volumes: # Source code (live mount — changes reflect immediately via uvicorn --reload) - ${SEFOLDER}/backend:/app/backend - ${SEFOLDER}/lyrics-sync:/app/lyrics-sync # Persistent storage for uploaded files and job outputs - ${SEFOLDER}/jobs:/app/media/jobs - ${SEFOLDER}/uploads:/app/media/uploads - ${SEFOLDER}/imports:/app/media/imports - ${SEFOLDER}/db:/app/media/db # LRC sync data - ${SEFOLDER}/lrc_sync:/app/media/lrc_sync # Model caches (avoid re-downloading after container restart) - ${SEFOLDER}/stemengine-models:/root/.cache - ${SEFOLDER}/opt:/opt # Projects folder (where separated stems are stored) - ${PROJECTS}:/app/media/projects command: python3.11 -m uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload environment: - PUID=${PUID:-1000} - PGID=${PGID:-1000} - TZ=${TZ:-UTC} # Open CORS mode: no origin restrictions. - CORS_ORIGINS=${CORS_ORIGINS:-*} - CORS_ORIGIN_REGEX=${CORS_ORIGIN_REGEX:-} - ADMIN_TOKEN=${ADMIN_TOKEN:?Set ADMIN_TOKEN to a strong non-empty secret} # Root media directory — all path defaults derive from this. - MEDIA_DIR=/app/media # Storage layout (defaults shown — uncomment and change to override) - UPLOAD_DIR=/app/media/uploads - JOBS_DIR=/app/media/jobs - PROJECTS_DIR=/app/media/projects - DB_PATH=/app/media/db/stemengine.db - LRC_SYNC_DIR=/app/media/lrc_sync - BS_ROFORMER_MODEL_DIR=/opt/bs_roformer_models - MEL_BAND_MODEL_DIR=/opt/mel_band_models - MDXNET_MODEL_DIR=/opt/mdxnet_models - SCNET_MODEL_DIR=/opt/scnet_models - MEDLEYVOX_MODEL_DIR=/opt/medleyvox_models - APOLLO_MODEL_DIR=/opt/apollo_models - ASPIRATION_MODEL_DIR=/opt/aspiration_models - BANDIT_MODEL_DIR=/opt/bandit_models - CPU_SEPARATION=${CPU_SEPARATION:-false} - STEMENGINE_GPU_POWER_LIMIT=${STEMENGINE_GPU_POWER_LIMIT:-100} # --------------------------------------------------------------------------- # RTX 3060 / constrained-VRAM stability # --------------------------------------------------------------------------- # When true (default): on CUDA OOM after all param reductions, only fall back # to a model that also produces lead/backing vocal stems (karaoke-capable). # Set false to raise immediately instead of attempting cross-model fallback. - FALLBACK_REQUIRE_LEAD_BACK=${FALLBACK_REQUIRE_LEAD_BACK:-true} # Tracks longer than this many seconds get MelBand eco params forced # (batch_size=1, num_overlap=2) regardless of active power preset. # Default 480 s = 8 min. Raise for more powerful GPUs. - LONG_TRACK_DURATION_SECS=${LONG_TRACK_DURATION_SECS:-480} # When true, MelBand jobs emit per-stage memory diagnostics so you can see # the cost of audio load, model load, preallocation, and cleanup. - MELBAND_STAGE_DIAGNOSTICS=${MELBAND_STAGE_DIAGNOSTICS:-false} # MelBand segment size (dim_t) override for all MelBand models. # Smaller values (e.g., 256) = faster inference, less fan noise # Larger values (e.g., 1101) = slower but potentially better quality # Default: unset (uses model's native config value) # Common values: 256 (Becruily/fast), 801 (standard), 1101 (Gabox/slow) - MELBAND_DIM_T_OVERRIDE=${MELBAND_DIM_T_OVERRIDE:-} # Gabox rollout controls - GABOX_ENABLE_SMALL_MODELS=${GABOX_ENABLE_SMALL_MODELS:-false} - GABOX_ENABLE_EXPERIMENTAL_MODELS=${GABOX_ENABLE_EXPERIMENTAL_MODELS:-false} - GABOX_SMALL_MODEL_MAX_DURATION_SECS=${GABOX_SMALL_MODEL_MAX_DURATION_SECS:-360} - GABOX_SMALL_MODEL_PREFERRED_FOR_LOW_VRAM=${GABOX_SMALL_MODEL_PREFERRED_FOR_LOW_VRAM:-false} - MAX_UPLOAD_MB=${MAX_UPLOAD_MB:-500} - PYTHONUNBUFFERED=1 # Prevent CUDA OOM from allocator fragmentation when multiple large models # (Demucs, BS-RoFormer, Whisper) are loaded in the same process. - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True # Secret key for encrypting credentials at rest (required for Settings UI storage) # Generate with: python3 -c "import secrets; print(secrets.token_hex(32))" - SECRET_KEY=${SECRET_KEY} # Subsonic server (optional — can also be set via the Settings UI) - SUBSONIC_URL=${NAVSERVER} - SUBSONIC_USERNAME=${NAVUSER} - SUBSONIC_PASSWORD=${NAVPASS} - HF_TOKEN=${HF_TOKEN} # Omnizart MIDI transcription (optional — set OMNIZART_ENABLED=true to activate) # Set to 'true' once the omnizart-sidecar service is running. - OMNIZART_ENABLED=${OMNIZART_ENABLED:-true} # URL of the omnizart-sidecar container (used when OMNIZART_ENABLED=true). # Must be reachable from inside the stemengine container. - OMNIZART_SIDECAR_URL=${OMNIZART_SIDECAR_URL:-http://omnizart-sidecar:9765} # Parakeet multilingual transcription (self-hosted sidecar) - PARAKEET_ENABLED=${PARAKEET_ENABLED:-false} - PARAKEET_SIDECAR_URL=${PARAKEET_SIDECAR_URL:-http://parakeet-sidecar:8777} - PARAKEET_HEALTH_PATH=${PARAKEET_HEALTH_PATH:-/health} - PARAKEET_TRANSCRIBE_PATH=${PARAKEET_TRANSCRIBE_PATH:-/transcribe} - PARAKEET_TIMEOUT_SEC=${PARAKEET_TIMEOUT_SEC:-180} - DEFAULT_PARAKEET_MODEL=${DEFAULT_PARAKEET_MODEL:-parakeet-1.1b-rnnt-multilingual-asr} # Parakeet multilingual transcription (self-hosted only) - PARAKEET_ENABLED=${PARAKEET_ENABLED:-false} - PARAKEET_SIDECAR_URL=${PARAKEET_SIDECAR_URL:-http://parakeet-sidecar:8777} - PARAKEET_HEALTH_PATH=${PARAKEET_HEALTH_PATH:-/health} - PARAKEET_TRANSCRIBE_PATH=${PARAKEET_TRANSCRIBE_PATH:-/transcribe} - PARAKEET_TIMEOUT_SEC=${PARAKEET_TIMEOUT_SEC:-180} - DEFAULT_PARAKEET_MODEL=${DEFAULT_PARAKEET_MODEL:-parakeet-1.1b-rnnt-multilingual-asr} deploy: resources: reservations: devices: - driver: nvidia count: all capabilities: [gpu] stemengine-webui: container_name: stemengine-webui image: stemengine-webui:latest build: context: ${SEFOLDER}/frontend dockerfile: Dockerfile.webui args: # Build-time default API endpoint used when no custom address is set in Settings. # Users can override this at runtime in Settings → Integrations. - VITE_STEMENGINE_API_BASE=${VITE_STEMENGINE_API_BASE:-http://localhost:8000} restart: unless-stopped depends_on: - stemengine networks: - t2_proxy ports: - "5173:80" # labels: # - traefik.enable=true # - traefik.docker.network=t2_proxy # - traefik.http.routers.stemengine-web.rule=Host(`${STEMENGINE_WEB_HOST:-stemengine.ts.net}`) # - traefik.http.routers.stemengine-web.entrypoints=${TRAEFIK_ENTRYPOINTS:-websecure} # - traefik.http.routers.stemengine-web.tls=true # - traefik.http.routers.stemengine-web.service=stemengine-web-svc # - traefik.http.services.stemengine-web-svc.loadbalancer.server.port=80 # ── Omnizart Sidecar ────────────────────────────────────────────────────── # Build: docker compose -f stemengine.yml build omnizart-sidecar # Start: docker compose -f stemengine.yml --profile omnizart up -d omnizart-sidecar: container_name: omnizart-sidecar image: omnizart-sidecar:latest build: context: ${SEFOLDER}/omnizart-sidecar dockerfile: Dockerfile restart: unless-stopped networks: - t2_proxy volumes: # Cache Omnizart models between restarts (~500 MB) - omnizart-model-cache:/root/.omnizart environment: - PYTHONUNBUFFERED=1 - OMNIZART_BIN=${OMNIZART_BIN:-omnizart} # Keep sidecar TensorFlow from consuming full VRAM. # Falls back to STEMENGINE_GPU_VRAM_RESERVE_MB when unset. - OMNIZART_GPU_VRAM_RESERVE_MB=${OMNIZART_GPU_VRAM_RESERVE_MB:-${STEMENGINE_GPU_VRAM_RESERVE_MB:-2048}} # No GPU reservation by default — Omnizart can run on CPU (slower). # Uncomment the block below to grant GPU access: deploy: resources: reservations: devices: - driver: nvidia count: 1 capabilities: [gpu] profiles: - omnizart # ── Parakeet Sidecar ────────────────────────────────────────────────────── # Build: docker compose -f stemengine.yml build parakeet-sidecar # Start: docker compose -f stemengine.yml --profile parakeet up -d parakeet-sidecar: container_name: parakeet-sidecar build: context: ${SEFOLDER}/parakeet-sidecar dockerfile: Dockerfile restart: unless-stopped networks: - t2_proxy environment: - PYTHONUNBUFFERED=1 - DEFAULT_PARAKEET_MODEL=${DEFAULT_PARAKEET_MODEL:-parakeet-1.1b-rnnt-multilingual-asr} deploy: resources: reservations: devices: - driver: nvidia count: 1 capabilities: [gpu] profiles: - parakeet # ── Backend test runner (targeted hardening regression profile) ─────────── # Usage: # docker compose -f stemengine.yml --profile tests run --rm stemengine-tests stemengine-tests: container_name: stemengine-tests image: stemengine:latest pull_policy: never entrypoint: [] command: >- bash -lc "python3.11 -m pip install --no-cache-dir -r /app/backend/requirements-test.txt >/tmp/stemengine-tests-pip.log 2>&1 && python3.11 -m unittest backend.tests.test_jobs_api backend.tests.test_project_cleanup backend.tests.test_projects_delete backend.tests.test_findings_3_4_unit backend.tests.test_ensemble_separator backend.tests.test_optional_hardening backend.tests.test_stem_finalize -v" environment: - TZ=${TZ:-UTC} - SECRET_KEY=${SECRET_KEY:-test-secret-key} - CPU_SEPARATION=true - PYTHONUNBUFFERED=1 volumes: - ./backend:/app/backend - ./lyrics-sync:/app/lyrics-sync profiles: - tests volumes: # Omnizart sidecar model cache — avoids re-downloading on restart. omnizart-model-cache: projects: sekfold: driver_opts: type: cifs o: username=$SMBUSER,password=$SMBPASS,uid=$PUID,gid=$PGID,vers=3.0 device: //MSI/Eye/Music/Organized/Karaoke/StemEngine networks: t2_proxy: external: true