| # CPU-only Hugging Face Space image. | |
| # Serving uses llama.cpp + a pre-merged GGUF (Q4_K_M), so there is no torch / | |
| # transformers / bitsandbytes in the production image. This keeps the image small, | |
| # cold starts fast, and RAM well under the 16 GB free-tier ceiling. | |
| FROM python:3.11-slim | |
| # System deps. build-essential + cmake are a safety net in case pip falls back to | |
| # building llama-cpp-python from source; normally the prebuilt CPU wheel is used. | |
| RUN apt-get update && apt-get install -y --no-install-recommends \ | |
| git \ | |
| curl \ | |
| build-essential \ | |
| cmake \ | |
| && rm -rf /var/lib/apt/lists/* | |
| WORKDIR /app | |
| # Build llama-cpp-python from source against THIS image's glibc toolchain. | |
| # Why not a prebuilt wheel: the abetlen CPU "wheel index" ships musl-linked wheels | |
| # for every version recent enough to support the phi3 architecture (>=0.2.63), which | |
| # fail to load on this Debian/glibc image with: | |
| # "libc.musl-x86_64.so.1: cannot open shared object file". | |
| # The only glibc wheels there (<=0.2.62) predate phi3 support. So we compile. | |
| # GGML_NATIVE=OFF -> portable CPU binary (the build host CPU may differ from the | |
| # cpu-basic runtime CPU; -march=native would risk an illegal-instruction crash). | |
| # Version note: 0.3.2's bundled llama.cpp SEGFAULTs (exit 139) at inference time on | |
| # this GGUF (produced by a much newer llama.cpp). 0.3.19 is verified to load and | |
| # generate correct SQL against the exact production GGUF, so we pin to it. | |
| ENV CMAKE_ARGS="-DGGML_NATIVE=OFF" | |
| # Compile ONE translation unit at a time. A parallel compile (default = all cores) | |
| # OOM-killed the build runner (exit 137). Single-threaded keeps peak RAM ~1-2 GB. | |
| # Slower build, but reliable on the memory-limited HF build runner. | |
| ENV CMAKE_BUILD_PARALLEL_LEVEL=1 | |
| RUN pip install --no-cache-dir --no-binary=llama-cpp-python "llama-cpp-python==0.3.19" | |
| # Install the slim CPU-serving dependencies. | |
| COPY requirements-space.txt . | |
| RUN pip install --no-cache-dir -r requirements-space.txt | |
| # Copy project source | |
| COPY . . | |
| # HF Spaces requires the app to listen on port 7860 | |
| EXPOSE 7860 | |
| # Create dirs that may not exist in the repo | |
| RUN mkdir -p data models templates static | |
| # HF Spaces runs containers as a non-root user | |
| RUN useradd -m -u 1000 appuser && chown -R appuser /app | |
| USER appuser | |
| # Point HF cache to a writable location inside the container (GGUF is cached here) | |
| ENV HF_HOME=/app/.cache/huggingface | |
| CMD ["python", "src/server.py"] | |