# ollama-gfx906/Dockerfile
#
# Custom ollama image with ROCm 6.1 + gfx906 (MI50) support.
# The default ollama/rocm image ships ROCm 7.2 which dropped gfx906 support.
# This builds ollama from source targeting AMDGPU_TARGETS=gfx906.
#
# Build: docker build -t ollama/ollama:rocm-gfx906 .

FROM rocm/dev-ubuntu-22.04:6.1.2-complete AS builder

RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
    git golang cmake build-essential pkg-config \
    && rm -rf /var/lib/apt/lists/*

ARG OLLAMA_VERSION=v0.13.5
RUN git clone --depth 1 --branch ${OLLAMA_VERSION} https://github.com/ollama/ollama.git /build
WORKDIR /build

ENV HIP_PATH=/opt/rocm
ENV ROCM_PATH=/opt/rocm
ENV PATH=/opt/rocm/bin:/opt/rocm/hip/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
RUN cd llama.cpp && \
    mkdir build && cd build && \
    cmake .. \
      -DLLAMA_HIPBLAS=ON \
      -DCMAKE_C_COMPILER=clang \
      -DCMAKE_CXX_COMPILER=clang++ \
      -DAMDGPU_TARGETS=gfx906 \
      -DCMAKE_BUILD_TYPE=Release \
      -DLLAMA_NATIVE=OFF \
      -DLLAMA_BUILD_TESTS=OFF \
      -DLLAMA_BUILD_EXAMPLES=OFF \
      -DLLAMA_BUILD_SERVER=OFF && \
    cmake --build . --config Release -j$(nproc) && \
    cmake --install . --prefix /build/dist

ENV CGO_ENABLED=0
RUN go build -trimpath -o dist/ollama .

FROM ubuntu:22.04

RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
    ca-certificates curl libstdc++6 libgomp1 \
    && rm -rf /var/lib/apt/lists/*

COPY --from=builder /opt/rocm/lib/ /opt/rocm/lib/
COPY --from=builder /opt/rocm/share/ /opt/rocm/share/
COPY --from=builder /build/dist/ollama /usr/bin/ollama
COPY --from=builder /build/dist/lib/ /usr/lib/ollama/

RUN ldconfig

ENV LD_LIBRARY_PATH=/opt/rocm/lib:/usr/lib/ollama
ENV HSA_OVERRIDE_GFX_VERSION=9.0.6

EXPOSE 11434
CMD ["serve"]