Dockerfile

FROM nvidia/cuda:12.9.0-runtime-ubuntu24.04

# Avoid interactive prompts during package installation
ENV DEBIAN_FRONTEND=noninteractive

# Install Python and system dependencies
# Ubuntu 24.04 ships FFmpeg 6.1 (torchcodec requires FFmpeg 5+, <8)
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 \
    python3-pip \
    python3-dev \
    git \
    libsndfile1 \
    ffmpeg \
    && rm -rf /var/lib/apt/lists/*

# Set Python alias (Ubuntu 24.04 ships Python 3.12)
RUN ln -sf /usr/bin/python3 /usr/bin/python

# Allow pip to install packages system-wide in the container (PEP 668)
ENV PIP_BREAK_SYSTEM_PACKAGES=1

# Set working directory
WORKDIR /app

# Install PyTorch ecosystem (cu128 wheels for CUDA 12.8+/12.9 compat)
RUN pip install --no-cache-dir \
    torch==2.8.0 \
    torchaudio==2.8.0 \
    torchcodec==0.6.0 \
    --index-url https://download.pytorch.org/whl/cu128

# Copy and install common requirements (torch already installed above, pip will skip it)
COPY requirements-docker.txt /app/requirements/requirements-docker.txt
RUN pip install --no-cache-dir -r /app/requirements/requirements-docker.txt

# Install transformers from source
RUN pip install --no-cache-dir git+https://github.com/huggingface/transformers.git

# Install additional dependencies for specific model families
# - Voxtral requires mistral-common[audio]
# - soundfile for audio I/O
RUN pip install --no-cache-dir \
    "mistral-common[audio]>=1.9.0" \
    soundfile

# Copy the TTS script last — it changes far more often than deps, so keeping
# it in its own layer lets every install layer above stay cached on rebuild.
COPY tts_script.py /app/tts_script.py
COPY eval.py /app/eval.py

# Default entrypoint
ENTRYPOINT ["bash"]

# Keep-alive CMD so the Space runtime stays healthy. HF Jobs and `docker run`
# override this with their own command (e.g. `run_cohere.sh`).
EXPOSE 7860
CMD ["-c", "python3 -m http.server 7860"]