Dockerfile

# nlp-asr-kit — Whisper small + Qwen2.5-0.5B-Instruct Q4 GGUF
# Optimized for Hugging Face Spaces (free CPU tier, glibc/Debian)

FROM python:3.10

# ffmpeg for audio decoding (faster-whisper)
# build-essential + cmake needed to compile llama-cpp-python for glibc
RUN apt-get update && apt-get install -y \
    ffmpeg \
    build-essential \
    cmake \
    && rm -rf /var/lib/apt/lists/*

RUN useradd -m -u 1000 user
USER user
ENV PATH="/home/user/.local/bin:$PATH"
ENV HF_HOME="/home/user/.cache/huggingface"

WORKDIR /app

COPY --chown=user ./requirements.txt requirements.txt

# Compile llama-cpp-python for CPU on glibc (Debian)
# MAKE_JOBS=1 limits parallelism to prevent OOMKilled during compilation
# This takes ~5 min but is the only reliable way on glibc containers
RUN CMAKE_ARGS="-DLLAMA_BLAS=OFF -DLLAMA_CUBLAS=OFF" \
    MAKE_JOBS=1 \
    pip install --no-cache-dir --user llama-cpp-python==0.2.90

RUN pip install --no-cache-dir --user -r requirements.txt

COPY --chown=user . /app

CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860"]