# OrpheusTail - Orpheus TTS Service for NVIDIA Jetson AGX Orin
# 
# Uses dustynv's vLLM container which has proper Jetson CUDA support
# Orpheus uses vLLM under the hood for fast inference

FROM dustynv/vllm:0.8.6-r36.4-cu128-24.04

# Set working directory
WORKDIR /app

# Install system dependencies
RUN apt-get update && apt-get install -y \
    git \
    wget \
    libsndfile1 \
    && rm -rf /var/lib/apt/lists/*

# Copy requirements first for better caching
COPY requirements.txt /app/

# Install Python dependencies (FastAPI, etc - but NOT torch/vllm)
RUN pip3 install --no-cache-dir -r requirements.txt

# Install orpheus-speech from GitHub repo (supports max_model_len) WITHOUT dependencies
# to avoid overwriting vllm/torch. Then install snac audio codec.
RUN pip3 install --no-cache-dir --no-deps git+https://github.com/canopyai/Orpheus-TTS.git#subdirectory=orpheus_tts_pypi && \
    pip3 install --no-cache-dir --index-url https://pypi.org/simple/ snac

# Copy application code
COPY main.py /app/

# Create directories for cache, output, and custom voices
RUN mkdir -p /app/cache /app/output /app/voices

# Expose API port (same as VoiceTail for drop-in replacement)
EXPOSE 8766

# Set environment variables
ENV PYTHONUNBUFFERED=1
ENV CACHE_DIR=/app/cache
ENV OUTPUT_DIR=/app/output
ENV VOICES_DIR=/app/voices
ENV ORPHEUS_MODEL=canopylabs/orpheus-tts-0.1-finetune-prod
ENV DEFAULT_VOICE=tara
ENV MAX_MODEL_LEN=2048

# Health check (longer start period - model loading takes time)
HEALTHCHECK --interval=30s --timeout=10s --start-period=180s --retries=3 \
    CMD curl -f http://localhost:8766/health || exit 1

# Run the FastAPI application
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8766"]