Dockerfile.rocm.ubi

## Global Args ##################################################################
ARG BASE_UBI_IMAGE_TAG
ARG PYTHON_VERSION

FROM registry.access.redhat.com/ubi9/ubi-minimal:${BASE_UBI_IMAGE_TAG} AS base

ARG PYTHON_VERSION

ENV VIRTUAL_ENV=/opt/vllm
ENV PATH="$VIRTUAL_ENV/bin:$PATH"

RUN --mount=type=cache,target=/root/.cache/pip \
 microdnf -y update && \
 microdnf install -y --setopt=install_weak_deps=0 --nodocs \
    python${PYTHON_VERSION}-devel \
    python${PYTHON_VERSION}-pip \
    python${PYTHON_VERSION}-wheel && \
    python${PYTHON_VERSION} -m venv $VIRTUAL_ENV && \
    pip install -U pip wheel setuptools uv && \
 microdnf clean all


FROM base AS rocm_base
ARG ROCM_VERSION
ARG AMDGPU_VERSION=${ROCM_VERSION}
ARG PYTHON_VERSION
ARG BASE_UBI_IMAGE_TAG

RUN printf "[amdgpu]\n\
name=amdgpu\n\
baseurl=https://repo.radeon.com/amdgpu/${AMDGPU_VERSION}/rhel/${BASE_UBI_IMAGE_TAG/-*/}/main/x86_64/\n\
enabled=1\n\
priority=50\n\
gpgcheck=1\n\
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key\n\
[ROCm-${ROCM_VERSION}]\n\
name=ROCm${ROCM_VERSION}\n\
baseurl=https://repo.radeon.com/rocm/rhel9/${ROCM_VERSION}/main\n\
enabled=1\n\
priority=50\n\
gpgcheck=1\n\
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key" > /etc/yum.repos.d/amdgpu.repo


RUN --mount=type=cache,target=/root/.cache/uv \
    microdnf install -y --nodocs \
        # aiter calls `hipconfig` on import, which requires hip-devel. It also requires `git` until https://github.com/ROCm/aiter/pull/2226 is merged
        git \
        amd-smi-lib \
        hipcc \
        hip-devel \
        # these -devel packages are required for aiter JIT
        hipsparse-devel rocthrust-devel hipblas-devel hipblaslt-devel hipsolver-devel \
        # Install libdrm-amdgpu to avoid errors when retrieving device information (amdgpu.ids: No such file or directory)
        libdrm-amdgpu \
        && \
    microdnf clean all

ENV PATH=/opt/rocm/bin:$PATH \
    ROCM_HOME=/opt/rocm \
    ROCM_PATH=/opt/rocm

##################################################################################################

FROM rocm_base AS vllm-openai
ARG PYTHON_VERSION

WORKDIR /workspace

ENV VIRTUAL_ENV=/opt/vllm
ENV PATH=$VIRTUAL_ENV/bin:$PATH

# Required for triton (gcc, rsync), audio multimodal support (libsndfile), aiter (g++)
RUN microdnf install -y --setopt=install_weak_deps=0 --nodocs \
    gcc g++ \
    rsync libsndfile && \
    microdnf clean all

RUN --mount=type=cache,target=/root/.cache/uv \
    --mount=type=bind,src=payload,target=/workspace/payload \
    ./payload/run.sh

ENV HF_HUB_OFFLINE=1 \
    VLLM_USAGE_STATS_SERVER=https://console.redhat.com/api/rhaiis-stats \
    VLLM_USAGE_SOURCE=quay-docker-image \
    VLLM_WORKER_MULTIPROC_METHOD=spawn \
    # Silences the HF Tokenizers warning
    TOKENIZERS_PARALLELISM=false  \
    RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1 \
    VLLM_ROCM_USE_AITER=1 \
    HIP_FORCE_DEV_KERNARG=1 \
    OUTLINES_CACHE_DIR=/tmp/outlines \
    NUMBA_CACHE_DIR=/tmp/numba \
    TRITON_CACHE_DIR=/tmp/triton \
    VLLM_CACHE_ROOT=/tmp/vllm \
    MIOPEN_CACHE_DIR=/tmp/miopen

# setup non-root user for OpenShift
RUN umask 002 && \
    useradd --uid 2000 --gid 0 vllm && \
    mkdir -p /licenses /home/vllm && \
    chmod g+rwx /home/vllm

# pre-download tiktoken/harmony tokenizer for disconnected environments
# openai_harmony uses TIKTOKEN_RS_CACHE_DIR
# while the python tiktoken module uses TIKTOKEN_CACHE_DIR
ENV TIKTOKEN_CACHE_DIR=/home/vllm/.cache/tiktoken
ENV TIKTOKEN_RS_CACHE_DIR=${TIKTOKEN_CACHE_DIR}
RUN --mount=type=bind,src=tools/predownload_tiktoken_tokenizers.py,target=predownload_tiktoken_tokenizers.py \
    python predownload_tiktoken_tokenizers.py && \
    chmod g+rwx "${TIKTOKEN_CACHE_DIR}" && \
    chown -R vllm:0 $(dirname ${TIKTOKEN_CACHE_DIR})

COPY LICENSE /licenses/vllm.md
COPY examples/*.jinja /app/data/template/
RUN mkdir -p /opt/app-root && \
    ln -s /app/data/template /opt/app-root/template

USER 2000
WORKDIR /home/vllm

ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]