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. git-lfs is needed to pull the model checkpoints.
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 \
    python3-pip \
    python3-dev \
    git \
    git-lfs \
    libsndfile1 \
    ffmpeg \
    && rm -rf /var/lib/apt/lists/* \
    && git lfs install

# 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

WORKDIR /app

# Install PyTorch (cu128 wheels for CUDA 12.8+/12.9 compat). Installed before the model's
# requirements.txt so its unpinned torch/torchaudio entries are already satisfied.
RUN pip install --no-cache-dir \
    torch==2.8.0 \
    torchaudio==2.8.0 \
    --index-url https://download.pytorch.org/whl/cu128

# Clone the Inflect-Nano-v1 model repo (with LFS weights) and install its requirements.
# It ships the `inference` module + vendored tiny_tts frontend that run_eval.py imports.
RUN git clone https://huggingface.co/owensong/Inflect-Nano-v1 /opt/Inflect-Nano-v1 \
    && pip install --no-cache-dir -r /opt/Inflect-Nano-v1/requirements.txt

# Pre-download the NLTK data g2p_en needs (used by the text frontend), so no runtime download.
RUN python3 -c "import nltk; [nltk.download(p) for p in ('averaged_perceptron_tagger', 'averaged_perceptron_tagger_eng', 'cmudict')]"

# datasets + tqdm for the eval loop. numba (alignment kernel) and inflect (text frontend)
# are imported by the vendored tiny_tts frontend but are missing from the model's requirements.txt.
RUN pip install --no-cache-dir datasets tqdm numba inflect

# Copy the full repository
COPY . /app

# Default entrypoint
ENTRYPOINT ["bash"]

# Keep-alive CMD so the Space runtime stays healthy; `docker run` overrides it.
EXPOSE 7860
CMD ["-c", "python3 -m http.server 7860"]