FROM nvidia/cuda:13.4.3-cudnn-devel-ubuntu26.04
ENV DEBIAN_FRONTEND=noninteractive
RUN rm -f /etc/dpkg/dpkg.cfg.d/excludes && \
    rm -f /etc/apt/apt.conf.d/docker-clean && \
    rm -f /etc/apt/apt.conf.d/01-nodoc && \
    rm -f /usr/bin/man && \
    dpkg-divert --quiet --remove /usr/bin/man

RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
    --mount=type=cache,target=/var/lib/apt,sharing=locked \
    apt update && apt-get dist-upgrade --autoremove -y && apt-get install -y \
    python3-pip \
    python3-dev \
    python3-venv \
    build-essential \
    man \
    man-db \
    wget \
    bzip2 \
    curl \
    git \
    cmake

RUN python3 -m venv /root/venv
ENV PATH="/root/venv/bin:${PATH}"
RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install torch torchvision torchaudio --upgrade --index-url \
    https://download.pytorch.org/whl/cu134/

WORKDIR /root
RUN mkdir llama.cpp
COPY --exclude=build/ llama.cpp ./llama.cpp/
WORKDIR /root/llama.cpp
ENV LD_LIBRARY_PATH=/usr/local/cuda-13.4/compat:${LD_LIBRARY_PATH}
ENV LIBRARY_PATH=/usr/local/cuda-13.4/compat:${LIBRARY_PATH}

RUN cmake -B build -S . \
     -DCMAKE_BUILD_TYPE=Release \
     -DGGML_CUDA=ON \
     -DCMAKE_CUDA_ARCHITECTURES=120 \
     -DLLAMA_BUILD_TESTS=OFF \
     -DLLAMA_BUILD_EXAMPLES=ON \
    && cmake --build build -j$(nproc) \
    && cmake --build build --target install \
    && rm -rf build

WORKDIR /root
RUN git clone https://github.com/silveroxides/convert_to_quant && cd convert_to_quant && pip3 install -r requirements.txt && pip3 install .

RUN mkdir nvfp4_gguf_quant
WORKDIR /root/nvfp4_gguf_quant
COPY --exclude=__pycache__ nvfp4gguf ./nvfp4gguf
COPY --chmod=0644 gguf_to_nvfp4_gguf.py hf_to_nvfp4_gguf.py hf_to_nvfp4_safetensors.py requirements.txt safetensors_to_nvfp4_gguf.py hf_info.py gguf_info.py .

RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install -r requirements.txt
COPY --chmod=0644 docker/README.txt /workspace/README.txt
COPY --chmod=0755 docker/hf_info docker/gguf_info docker/hf_to_nvfp4_gguf docker/hf_to_nvfp4_safetensors docker/gguf_to_nvfp4_gguf docker/safetensors_to_nvfp4_gguf /usr/local/bin/

RUN mkdir -p /usr/local/share/man/man1
COPY --chmod=0644 man/man1/* /usr/local/share/man/man1/

RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install heretic-llm
RUN rm -f /opt/nvidia/entrypoint.d/*banner*.sh /opt/nvidia/entrypoint.d/*license*.sh 2>/dev/null

WORKDIR /workspace