FROM nvidia/cuda:12.1.1-runtime-ubuntu22.04

# Avoid interactive prompts during package installation
ENV DEBIAN_FRONTEND=noninteractive

# Install Python 3.10 and system dependencies
# SpeechBrain requires Python 3.10 and specific torch/transformers versions
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3.10 \
    python3.10-dev \
    python3.10-venv \
    python3-pip \
    git \
    libsndfile1 \
    ffmpeg \
    && rm -rf /var/lib/apt/lists/*

# Set Python 3.10 as default
RUN ln -sf /usr/bin/python3.10 /usr/bin/python3 && \
    ln -sf /usr/bin/python3 /usr/bin/python

# Allow pip to install packages system-wide
ENV PIP_BREAK_SYSTEM_PACKAGES=1

# Set working directory
WORKDIR /app

# Install PyTorch ecosystem (cu121 wheels to match CUDA 12.1 base)
RUN pip install --no-cache-dir \
    torch==2.5.1 \
    torchaudio==2.5.1 \
    torchvision==0.20.1 \
    --index-url https://download.pytorch.org/whl/cu121

# Pin datasets version for SpeechBrain compatibility
RUN pip install --no-cache-dir datasets==2.19.1

# Install common requirements (torch already installed above, pip will skip it)
RUN pip install --no-cache-dir \
    evaluate \
    datasets \
    librosa \
    jiwer \
    num2words \
    peft

# Install SpeechBrain-specific dependencies (pinned transformers/accelerate)
RUN pip install --no-cache-dir \
    transformers==4.41.2 \
    accelerate==0.30.1 \
    speechbrain

# Force soundfile backend for datasets audio decoding (avoids torchcodec/FFmpeg issues)
ENV HF_AUDIO_DECODER_BACKEND=soundfile

# Copy the full repository
COPY . /app

# 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_parakeet.sh`).
EXPOSE 7860
CMD ["-c", "python3 -m http.server 7860"]