text-generation-inference/Dockerfile_llamacpp

ARG llama_version=b4599
ARG llama_hardware_target=cpu

FROM nvidia/cuda:12.6.3-cudnn-devel-ubuntu24.04 AS base

ENV DEBIAN_FRONTEND=noninteractive
RUN apt update && apt install -y \
    python3-venv \
    python3-pip

RUN python3 -m venv /venv
ENV PATH="/venv/bin:$PATH"
RUN pip3 install --no-cache-dir transformers

FROM base AS deps
WORKDIR /opt/src

RUN apt install -y \
    clang \
    cmake \
    curl \
    git \
    libssl-dev \
    pkg-config \
    tar

FROM deps AS llamacpp-builder
ARG llama_version
ENV LLAMA_VERSION=${llama_version}

ADD https://github.com/ggerganov/llama.cpp/archive/refs/tags/${LLAMA_VERSION}.tar.gz /opt/src/
RUN tar -xzf ${LLAMA_VERSION}.tar.gz && \
    cd llama.cpp-${LLAMA_VERSION} && \
    cmake \
    -B build \
    -DCMAKE_INSTALL_PREFIX=/usr/llama \
    -DCMAKE_C_COMPILER=clang \
    -DCMAKE_CXX_COMPILER=clang++ \
    -DGGML_CUDA=1 \
    -DLLAMA_BUILD_COMMON=OFF \
    -DLLAMA_BUILD_TESTS=OFF \
    -DLLAMA_BUILD_EXAMPLES=OFF \
    -DLLAMA_BUILD_SERVER=OFF \
 && cmake --build build --parallel --config Release -j \
 && cmake --install build

FROM deps AS rust-builder
COPY rust-toolchain.toml rust-toolchain.toml
RUN curl -sSf https://sh.rustup.rs | sh -s -- -y --no-modify-path --default-toolchain none
ENV PATH="/root/.cargo/bin:$PATH"

COPY . .
COPY --from=llamacpp-builder /usr/llama/lib/ /usr/lib/
COPY --from=llamacpp-builder /usr/llama/include/ /usr/include/


ARG llama_hardware_target
ENV TGI_LLAMA_HARDWARE_TARGET=${llama_hardware_target}
RUN export TGI_LIB_SEARCH_PATH=/usr/local/cuda/lib64:/usr/local/cuda/lib64/stubs && \
    ln -s /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1 && \
    cargo build \
    --profile release-opt \
    --package text-generation-router-llamacpp

FROM nvidia/cuda:12.6.3-cudnn-runtime-ubuntu24.04
WORKDIR /usr/bin

ENV DEBIAN_FRONTEND=noninteractive
ENV PATH="/venv/bin:$PATH"

RUN apt update && apt install -y \
    openssl \
    python3-venv \
    python3-pip

RUN python3 -m venv /venv && \
    pip3 install --no-cache-dir -r backends/llamacpp/requirements.txt

COPY --from=llamacpp-builder /usr/llama/lib/ /usr/lib/
COPY --from=llamacpp-builder /usr/llama/include/ /usr/include/
COPY --from=llamacpp-builder /usr/llama/bin/ /usr/bin/
COPY --from=rust-builder /opt/src/target/release-opt/text-generation-router-llamacpp /usr/bin/text-generation-launcher

ENTRYPOINT ["text-generation-launcher"]