Dockerfile

# Hugging Face Space (Docker) — GenSearcher + FireRed
# Requires GPU. For multi-GPU full-local mode, set START_VLLM_*=1 and CUDA device envs in README.

FROM pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime

ENV DEBIAN_FRONTEND=noninteractive
RUN apt-get update && apt-get install -y --no-install-recommends \
    curl \
    git \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

COPY vendor/rllm /app/vendor/rllm
COPY requirements.txt /app/requirements.txt
COPY app.py space_gen.py space_health.py /app/
COPY services /app/services
COPY scripts/entrypoint_body.sh scripts/verify_env.py /app/scripts/
# HF may exec /app/scripts/entrypoint.sh directly — never COPY it from Windows (CRLF → "exec format error").
# Build a 2-line LF-only stub; normalize body script bytes inside Linux.
RUN python3 -c "import pathlib; p=pathlib.Path('/app/scripts/entrypoint_body.sh'); b=p.read_bytes(); b=b.lstrip(b'\xef\xbb\xbf'); b=b.replace(b'\r\n', b'\n').replace(b'\r', b''); p.write_bytes(b)" \
    && printf '%s\n' '#!/bin/bash' 'exec /bin/bash /app/scripts/entrypoint_body.sh' > /app/scripts/entrypoint.sh \
    && chmod +x /app/scripts/entrypoint.sh /app/scripts/entrypoint_body.sh

ENV PYTHONPATH=/app/vendor/rllm
ENV GRADIO_SERVER_PORT=7860
# HF Spaces / minimal images often have uid 1000 with no /etc/passwd entry; PyTorch Inductor calls
# getpass.getuser() and crashes with KeyError. USER/LOGNAME short-circuit getuser(); cache dirs avoid $HOME issues.
ENV USER=huggingface
ENV LOGNAME=huggingface
ENV TORCHINDUCTOR_CACHE_DIR=/tmp/torch_inductor_cache
ENV TRITON_CACHE_DIR=/tmp/triton_cache

RUN pip install --no-cache-dir --upgrade pip setuptools wheel \
    && pip install --no-cache-dir -e /app/vendor/rllm \
    && pip install --no-cache-dir -r /app/requirements.txt

# Optional: local vLLM inside the image (large). Disable with build-arg if you only use external APIs.
ARG INSTALL_VLLM=1
RUN if [ "$INSTALL_VLLM" = "1" ]; then pip install --no-cache-dir "vllm>=0.6.3"; fi

EXPOSE 7860

# PID 1 is bash (real ELF). HF may ignore this and exec entrypoint.sh; that file is generated above with valid shebang.
CMD ["/bin/bash", "/app/scripts/entrypoint_body.sh"]