# ============================================================ # Single stage: Python runtime + Node.js # ============================================================ # CUDA 12.4 + cuDNN 9 runtime — provides GPU support for: # • PyTorch (cu124 wheels bundle their own cuBLAS/cuDNN) # • onnxruntime-gpu (requires system libcudnn; uses CUDAExecutionProvider for # MDX-Net separation and basic-pitch ONNX inference) # • libcudnn8 is also installed alongside cuDNN 9 because WhisperX's phoneme # alignment model (wav2vec2 via pyannote / CTranslate2) links against the # cuDNN 8 ABI (.so.8 filenames) which cuDNN 9 no longer ships. # # Node.js is also installed here so that start.sh can run # `vite build --watch`, keeping the frontend live without a Docker rebuild. # node_modules are installed once at first container start and persisted in # the stemengine-node-modules named volume. FROM nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04 ENV DEBIAN_FRONTEND=noninteractive # ---------- System dependencies ---------- # Add deadsnakes PPA for Python 3.11 (not available in default Ubuntu 22.04 repos). RUN apt-get update && apt-get install -y --no-install-recommends \ ca-certificates \ curl \ gnupg \ && mkdir -p /etc/apt/keyrings \ && curl -fsSL "https://keyserver.ubuntu.com/pks/lookup?op=get&search=0xF23C5A6CF475977595C89F51BA6932366A755776" \ | gpg --dearmor -o /etc/apt/keyrings/deadsnakes.gpg \ && echo "deb [signed-by=/etc/apt/keyrings/deadsnakes.gpg] https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu jammy main" \ > /etc/apt/sources.list.d/deadsnakes.list \ && apt-get update \ && apt-get install -y --no-install-recommends \ ffmpeg \ abcmidi \ libcudnn8 \ python3.11 \ python3.11-dev \ python3.11-venv \ build-essential \ git \ libsndfile1 \ libsndfile1-dev \ espeak-ng \ libespeak-ng-dev \ libmecab-dev \ && rm -rf /var/lib/apt/lists/* # Bootstrap pip for Python 3.11, then install uv (faster resolver, no depth limit) RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.11 \ && python3.11 -m pip install --no-cache-dir uv # ---------- Node.js 20.x ---------- # Required for `vite build --watch` which runs at container startup via start.sh. RUN curl -fsSL https://deb.nodesource.com/setup_20.x | bash - \ && apt-get install -y --no-install-recommends nodejs \ && rm -rf /var/lib/apt/lists/* # ---------- Main Python 3.11 dependencies ---------- WORKDIR /app COPY backend/requirements.txt /app/backend/requirements.txt RUN python3.11 -m uv pip install --system --no-cache -r /app/backend/requirements.txt # Install CUDA-enabled PyTorch (CUDA 12.4 wheels; self-contained, does not need system CUDA 12.4) # torch 2.6+ is required — earlier versions have a serious torch.load vulnerability (even with # weights_only=True). cu121 wheels were discontinued after torch 2.4; use cu124 going forward. RUN python3.11 -m uv pip install --system --no-cache \ torch==2.6.0 torchaudio==2.6.0 torchvision==0.21.0 \ --index-url https://download.pytorch.org/whl/cu124 # Install basic-pitch without its tensorflow dependency. # basic-pitch can run via onnxruntime (already installed above); installing with # --no-deps skips the tensorflow pull that has no cp311 wheels and conflicts with # numpy>=2.1.0 required by whisperx. RUN python3.11 -m uv pip install --system --no-cache --no-deps basic-pitch # Note: Model weights are downloaded on-demand at runtime on first use. # This keeps the Docker image smaller and avoids build-time network dependencies. # ---------- Application source ---------- COPY backend/ /app/backend/ COPY lyrics-sync/ /app/lyrics-sync/ # Frontend source is bind-mounted at runtime (${SEFOLDER}/frontend:/app/frontend). # Copy it here so the image has a fallback if the bind-mount is absent, # and so COPY frontend/package*.json is available for the layer cache. COPY frontend/ /app/frontend/ # ---------- Startup script ---------- COPY start.sh /app/start.sh RUN chmod +x /app/start.sh # ---------- Runtime ---------- ENV PYTHONUNBUFFERED=1 ENV PYTHONDONTWRITEBYTECODE=1 ENV MEDIA_DIR=/app/media ENV CPU_SEPARATION=true ENV BS_ROFORMER_MODEL_DIR=/opt/bs_roformer_models ENV MEL_BAND_MODEL_DIR=/opt/mel_band_models ENV MDXNET_MODEL_DIR=/opt/mdxnet_models ENV SCNET_MODEL_DIR=/opt/scnet_models ENV MEDLEYVOX_MODEL_DIR=/opt/medleyvox_models ENV APOLLO_MODEL_DIR=/opt/apollo_models ENV ASPIRATION_MODEL_DIR=/opt/aspiration_models ENV BANDIT_MODEL_DIR=/opt/bandit_models EXPOSE 8000 CMD ["/app/start.sh"]