Dockerfile

FROM pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PERSIST_ROOT=/app/data-v7 \
    HF_HOME=/app/data-v7/cache/huggingface \
    TRANSFORMERS_CACHE=/app/data-v7/cache/huggingface \
    HF_HUB_ENABLE_HF_TRANSFER=0 \
    TOKENIZERS_PARALLELISM=false \
    PORT=7860

RUN apt-get update && apt-get install -y --no-install-recommends \
      build-essential ca-certificates cmake curl git wget && \
    rm -rf /var/lib/apt/lists/*

WORKDIR /app
COPY requirements.txt .
RUN python -m pip install --no-cache-dir --upgrade pip && \
    python -m pip install --no-cache-dir -r requirements.txt

# Cache the public base model while the Space image is building. Hugging Face
# does not bill requested GPU hardware during the build stage.
RUN python -c "from huggingface_hub import snapshot_download; snapshot_download('Qwen/Qwen3-1.7B')"

# Pinned official llama.cpp revision for reproducible GGUF conversion.
RUN git clone https://github.com/ggml-org/llama.cpp.git /opt/llama.cpp && \
    cd /opt/llama.cpp && \
    git checkout 4762ad7316dcdec20016ab5985fb46a27902204d && \
    cmake -B build -DGGML_CUDA=OFF -DLLAMA_CURL=OFF -DGGML_NATIVE=OFF \
      -DGGML_AVX512=OFF -DGGML_AVX512_VBMI=OFF -DGGML_AVX512_VNNI=OFF \
      -DGGML_AVX512_BF16=OFF -DCMAKE_BUILD_TYPE=Release && \
    cmake --build build --config Release -j2 --target llama-quantize llama-server

COPY . /app
EXPOSE 7860
CMD ["python", "app.py"]