mirror of
https://github.com/modelscope/modelscope.git
synced 2026-07-10 04:22:33 +02:00
73 lines
2.4 KiB
Docker
73 lines
2.4 KiB
Docker
ARG BUILD_BASE_IMAGE=mx-devops-acr-cn-shanghai.cr.volces.com/opensource/public-ai-release/maca/ms-swift:3.10.3-maca.ai3.3.0.16-torch2.6-py310-ubuntu22.04-amd64
|
|
ARG PYTHON_VERSION=3.10
|
|
|
|
FROM ${BUILD_BASE_IMAGE} AS base
|
|
|
|
|
|
# may need passing a particular vllm version during build
|
|
ARG VLLM_VERSION
|
|
ARG VLLM_METAX_VERSION
|
|
ARG MEGATRON_VERSION
|
|
ARG SWIFT_VERSION
|
|
|
|
# --- 设置环境变量（可被 --build-arg 覆盖）---
|
|
ENV MACA_PATH=/opt/maca
|
|
ENV CUCC_CMAKE_ENTRY=2
|
|
ENV CUDA_PATH=/root/cu-bridge/CUDA_DIR
|
|
ENV CUCC_PATH=${MACA_PATH}/tools/cu-bridge
|
|
ENV PATH=/opt/conda/bin:/opt/conda/condabin:${CUDA_PATH}/bin:${CUCC_PATH}/tools:${CUCC_PATH}/bin:${MACA_PATH}/bin:${PATH}
|
|
ENV LD_LIBRARY_PATH=${CUDA_PATH}/lib64:${MACA_PATH}/lib:${MACA_PATH}/mxgpu_llvm/lib:${LD_LIBRARY_PATH}
|
|
RUN echo $PATH
|
|
RUN apt install -y git
|
|
# 检查并初始化 cu-bridge
|
|
RUN if [ ! -d /root/cu-bridge ]; then \
|
|
${MACA_PATH}/tools/cu-bridge/tools/pre_make; \
|
|
fi
|
|
|
|
# ======================
|
|
# Step 1: Clone and build original vLLM (for torch setup)
|
|
# ======================
|
|
WORKDIR /workspace
|
|
RUN git clone --branch ${VLLM_VERSION} https://github.com/vllm-project/vllm.git \
|
|
&& cd vllm \
|
|
&& python3 use_existing_torch.py \
|
|
&& pip install -r requirements/build.txt
|
|
|
|
# ======================
|
|
# Step 2: Build vLLM with empty device (to avoid CUDA dependency)
|
|
# ======================
|
|
RUN cd vllm \
|
|
&& VLLM_TARGET_DEVICE=empty pip install -v . --no-build-isolation
|
|
|
|
|
|
# ======================
|
|
# Step 3: Build vLLM-metax
|
|
# ======================
|
|
RUN git clone --branch ${VLLM_METAX_VERSION} https://github.com/MetaX-MACA/vLLM-metax.git \
|
|
&& cd vLLM-metax \
|
|
&& python3 use_existing_metax.py \
|
|
&& pip install -r requirements/build.txt \
|
|
&& python3 -m build -w -n \
|
|
&& pip install dist/*.whl
|
|
|
|
# ======================
|
|
# Step 4: Clone and patch Megatron-LM
|
|
# ======================
|
|
RUN git clone --branch ${MEGATRON_VERSION} https://github.com/NVIDIA/Megatron-LM.git \
|
|
&& sed -i 's/nvcc/cucc/g' /workspace/Megatron-LM/megatron/legacy/fused_kernels/__init__.py
|
|
|
|
# ======================
|
|
# Step 5: Clone, patch and install ms-swift
|
|
# ======================
|
|
RUN rm -rf /workspace/ms-swift
|
|
|
|
RUN git clone --branch ${SWIFT_VERSION} https://github.com/modelscope/ms-swift.git \
|
|
&& sed -i '0,/^\(from \|import \)/{s//import vllm_metax.patch\n&/}' ms-swift/swift/__init__.py \
|
|
&& cd ms-swift \
|
|
&& pip install -r requirements.txt \
|
|
&& pip install .
|
|
|
|
|
|
# 默认命令
|
|
CMD ["bash"]
|