FROM nvidia/cuda:12.8.0-cudnn-devel-ubuntu24.04

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update &&     apt-get install -y --no-install-recommends --fix-missing     git git-lfs wget curl cmake build-essential     libssl-dev zlib1g-dev libbz2-dev libreadline-dev     libsqlite3-dev libncursesw5-dev xz-utils tk-dev     libxml2-dev libxmlsec1-dev libffi-dev liblzma-dev     python3 python3-pip python3-venv python-is-python3 ffmpeg  && rm -rf /var/lib/apt/lists/*

RUN id -u 1000 >/dev/null 2>&1 || useradd -m -u 1000 user

ENV HOME=/home/user
ENV VIRTUAL_ENV=/home/user/venv
ENV PATH="/home/user/venv/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:${PATH}"
ENV PYTHONPATH=/home/user/app
ENV PYTHONUNBUFFERED=1
ENV HF_HUB_ENABLE_HF_TRANSFER=1
ENV HF_HOME=/data/.huggingface
ENV HF_HUB_CACHE=/data/.huggingface/hub
ENV TRANSFORMERS_CACHE=/data/.huggingface/transformers
ENV TMPDIR=/tmp
ENV GRADIO_ALLOW_FLAGGING=never
ENV GRADIO_NUM_PORTS=1
ENV GRADIO_SERVER_NAME=0.0.0.0
ENV GRADIO_SERVER_PORT=7860
ENV GRADIO_THEME=huggingface
ENV TQDM_POSITION=-1
ENV TQDM_MININTERVAL=1
ENV SYSTEM=spaces
ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}

WORKDIR /home/user/app

RUN python -m venv /home/user/venv &&     pip install --no-cache-dir -U pip setuptools wheel &&     pip install --no-cache-dir       huggingface-hub       hf-transfer       "gradio[oauth]"       gradio_huggingfacehub_search       APScheduler

COPY --chown=1000 . /home/user/app

# Clone and build llama.cpp at image build time (CPU-only, no CUDA needed for quantization).
# This bakes the binaries into the image so startup is instant.
RUN rm -rf /home/user/app/llama.cpp &&     git clone https://github.com/ggerganov/llama.cpp /home/user/app/llama.cpp &&     pip install --no-cache-dir -r /home/user/app/llama.cpp/requirements/requirements-convert_hf_to_gguf.txt &&     cp /home/user/app/groups_merged.txt /home/user/app/llama.cpp/ &&     cd /home/user/app/llama.cpp &&     cmake -B build -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=OFF -DLLAMA_CURL=OFF &&     cmake --build build --config Release -j$(nproc) --target llama-quantize llama-gguf-split llama-imatrix &&     cp build/bin/llama-quantize . &&     cp build/bin/llama-gguf-split . &&     cp build/bin/llama-imatrix . &&     chmod +x ./llama-quantize ./llama-gguf-split ./llama-imatrix &&     rm -rf build

RUN chown -R 1000:1000 /home/user

USER 1000

EXPOSE 7860

CMD ["bash", "start.sh"]