Dockerfile

# Dockerfile of qwenllm/qwen-omni:2.5-cu121

ARG CUDA_VERSION=12.1.0
ARG from=nvidia/cuda:${CUDA_VERSION}-cudnn8-devel-ubuntu22.04

FROM ${from} as base

ARG DEBIAN_FRONTEND=noninteractive
RUN <<EOF
apt update -y && apt upgrade -y && apt install -y --no-install-recommends  \
    git \
    git-lfs \
    python3 \
    python3-pip \
    python3-dev \
    wget \
    vim \
    libsndfile1 \
    ccache \
    software-properties-common \
    ffmpeg \
&& rm -rf /var/lib/apt/lists/*
EOF

RUN wget https://github.com/Kitware/CMake/releases/download/v3.26.1/cmake-3.26.1-Linux-x86_64.sh \
    -q -O /tmp/cmake-install.sh \
    && chmod u+x /tmp/cmake-install.sh \
    && mkdir /opt/cmake-3.26.1 \
    && /tmp/cmake-install.sh --skip-license --prefix=/opt/cmake-3.26.1 \
    && rm /tmp/cmake-install.sh \
    && ln -s /opt/cmake-3.26.1/bin/* /usr/local/bin

RUN ln -s /usr/bin/python3 /usr/bin/python

RUN git lfs install

FROM base as dev

WORKDIR /

RUN mkdir -p /data/shared/Qwen

WORKDIR /data/shared/Qwen/

FROM dev as bundle_req
RUN --mount=type=cache,target=/root/.cache/pip pip3 install networkx==3.1
RUN --mount=type=cache,target=/root/.cache/pip pip3 install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 xformers==0.0.29.post2
RUN --mount=type=cache,target=/root/.cache/pip pip3 install git+https://github.com/huggingface/transformers@3a1ead0aabed473eafe527915eea8c197d424356  \
    && pip3 install accelerate qwen-omni-utils modelscope_studio

FROM bundle_req as bundle_vllm

ARG BUNDLE_FLASH_ATTENTION=true

ENV MAX_JOBS=8
ENV NVCC_THREADS=1
ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 8.9 9.0+PTX"
ENV VLLM_FA_CMAKE_GPU_ARCHES="80-real;90-real"
ENV CCACHE_DIR=/root/.cache/ccache

RUN --mount=type=cache,target=/root/.cache/ccache \
    --mount=type=cache,target=/root/.cache/pip \
    if [ "$BUNDLE_FLASH_ATTENTION" = "true" ]; then \
        mkdir -p /data/shared/code \
        && pip install ninja \
        && cd /data/shared/code \
        && git clone https://github.com/Dao-AILab/flash-attention.git \
        && cd flash-attention \
        && python setup.py install \
        && cd /data/shared/Qwen \
        && rm -rf /data/shared/code/flash-attention; \
    fi

ARG BUNDLE_VLLM=true

RUN --mount=type=cache,target=/root/.cache/ccache \
    --mount=type=cache,target=/root/.cache/pip \
    if [ "$BUNDLE_VLLM" = "true" ]; then \
    mkdir -p /data/shared/code \
        && cd /data/shared/code \
        && git clone -b qwen2_omni_public_v1 https://github.com/fyabc/vllm.git \
        && cd vllm \
        && python3 use_existing_torch.py \
        && pip3 install setuptools_scm \
        && pip3 install -r requirements/cuda.txt \
        && pip3 install . --no-build-isolation\
        && cd /data/shared/Qwen \
        && rm -rf /data/shared/code/vllm; \
    fi

RUN --mount=type=cache,target=/root/.cache/pip \
    pip3 install \
    gradio==5.21.0 \
    gradio_client==1.7.2 \
    librosa==0.11.0 \
    ffmpeg==1.4 \
    ffmpeg-python==0.2.0 \
    soundfile==0.13.1 \
    av

RUN rm -rvf /root/.cache/pip

COPY server.py ./
 
# Expose port 8000 (matching the port used in server.py)
EXPOSE 8000
 
# Set default command to run your server
CMD ["python3", "server.py"]