Dockerfile

FROM python:3.12-slim

ENV PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    PORT=7860 \
    HOME=/home/user \
    HF_HOME=/home/user/.cache/huggingface

WORKDIR /app
COPY requirements.txt ./
# The CPU wheel index prevents a CUDA runtime from being included in this
# CPU-only Space.  The qwen-asr 0.0.6 transformer backend is vendored under
# server_runtime, avoiding its unrelated ASR UI and alignment dependencies.
RUN pip install --no-cache-dir --index-url https://download.pytorch.org/whl/cpu torch==2.9.1 && \
    pip install --no-cache-dir -r requirements.txt
# Pin both remote bases at build time. Runtime loading names exact revisions,
# so a request cannot silently move to a different model version.
RUN python -c "from huggingface_hub import snapshot_download; snapshot_download('Qwen/Qwen3-ASR-0.6B', revision='5eb144179a02acc5e5ba31e748d22b0cf3e303b0'); snapshot_download('laion/clap-htsat-unfused', revision='8fa0f1c6d0433df6e97c127f64b2a1d6c0dcda8a')"
RUN useradd --create-home --uid 1000 user && \
    mkdir -p /home/user/.cache/huggingface && \
    chown -R user:user /home/user
COPY --chown=1000:1000 . .

USER user
EXPOSE 7860
CMD ["sh", "-c", "uvicorn server:app --host 0.0.0.0 --port ${PORT} --no-access-log"]