From 6f3200dfc277c03817c80c70178a410e350ebe2a Mon Sep 17 00:00:00 2001 From: baymax591 Date: Wed, 26 Nov 2025 09:56:47 +0800 Subject: [PATCH] [docker] Add Ascend NPU dockerfile (#1550) Co-authored-by: ji-huazhong Co-authored-by: vx120 <893600387@qq.com> --- docker/Dockerfile.ascend | 120 +++++++++++++++++++++++++++++++++++++++ docker/build_image.py | 34 +++++++++++ 2 files changed, 154 insertions(+) create mode 100644 docker/Dockerfile.ascend diff --git a/docker/Dockerfile.ascend b/docker/Dockerfile.ascend new file mode 100644 index 00000000..ce8db144 --- /dev/null +++ b/docker/Dockerfile.ascend @@ -0,0 +1,120 @@ +FROM {base_image} + +RUN pip config set global.index-url https://mirrors.aliyun.com/pypi/simple && \ + pip config set install.trusted-host mirrors.aliyun.com + +# Prepare required system dependencies +RUN apt-get update -y && \ + apt-get install -y --no-install-recommends gcc g++ cmake libnuma-dev wget git curl jq vim build-essential && \ + apt-get clean && \ + rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \ + pip install --upgrade pip setuptools packaging && \ + pip cache purge + +{extra_content} + +# Prepare repositories with low update frequency +RUN ARCH=$(uname -m) && \ + # Set extra pip index for x86_64 platform + echo "[LOG INFO] Detected architecture: $ARCH" && \ + if [ "$ARCH" = "x86_64" ]; then \ + pip config set global.extra-index-url "https://download.pytorch.org/whl/cpu/"; \ + fi && \ + # Clone libs + git clone --depth 1 --branch v0.11.0 https://github.com/vllm-project/vllm && \ + git clone --depth 1 --branch v0.11.0rc1 https://github.com/vllm-project/vllm-ascend.git && \ + git clone https://gitcode.com/Ascend/MindSpeed.git && \ + cd MindSpeed && git checkout f2b0977e && cd .. && \ + git clone --depth 1 --branch core_v0.12.1 https://github.com/NVIDIA/Megatron-LM.git + +# Install repositories with low update frequency +RUN ARCH=$(uname -m) && \ + # Export and source env + if [ "$ARCH" = "aarch64" ]; then \ + export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/8.3.RC1/aarch64-linux/devlib/linux/aarch64:$LD_LIBRARY_PATH; \ + elif [ "$ARCH" = "x86_64" ]; then \ + export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/8.3.RC1/x86_64-linux/devlib/linux/x86_64/:$LD_LIBRARY_PATH; \ + fi && \ + source /usr/local/Ascend/ascend-toolkit/set_env.sh && \ + source /usr/local/Ascend/nnal/atb/set_env.sh && \ + # Install torch & torch_npu & torchvision + pip install torch==2.7.1 torch_npu==2.7.1 torchvision==0.22.1 && \ + # Install vllm + cd vllm && VLLM_TARGET_DEVICE=empty pip install -v -e . && cd .. && \ + # Install vllm-ascend + cd vllm-ascend && pip install -v -e . && cd .. && \ + # Install MindSpeed & Megatron + pip install -e MindSpeed && \ + # Clear extra files + rm -rf /tmp/* /var/tmp/* && \ + pip cache purge + +ENV PYTHONPATH ${PYTHONPATH}:/Megatron-LM + +ARG INSTALL_MS_DEPS={install_ms_deps} + +# install dependencies +COPY requirements /var/modelscope + +RUN pip uninstall ms-swift modelscope -y && pip --no-cache-dir install pip==23.* -U && \ +if [ "$INSTALL_MS_DEPS" = "True" ]; then \ + pip --no-cache-dir install omegaconf==2.0.6 && \ + pip install 'editdistance==0.8.1' && \ + pip install --no-cache-dir 'cython<=0.29.36' versioneer 'numpy<2.0' -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/framework.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/audio.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/cv.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/multi-modal.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/nlp.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/science.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/tests.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir -r /var/modelscope/server.txt && \ + pip install --no-cache-dir https://modelscope.oss-cn-beijing.aliyuncs.com/packages/imageio_ffmpeg-0.4.9-py3-none-any.whl --no-dependencies --force && \ + pip install adaseq pai-easycv && \ + pip install --no-cache-dir 'scipy<1.13.0' && \ + pip install --no-cache-dir funtextprocessing typeguard==2.13.3 scikit-learn -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip install --no-cache-dir text2sql_lgesql==1.3.0 git+https://github.com/jin-s13/xtcocoapi.git@v1.14 git+https://github.com/gatagat/lap.git@v0.4.0 -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html --force --no-deps && \ + pip install --no-cache-dir mmcls>=0.21.0 mmdet>=2.25.0 decord>=0.6.0 mpi4py paint_ldm ipykernel fasttext -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip uninstall ddpm_guided_diffusion -y && \ + pip install --no-cache-dir 'blobfile>=1.0.5' && \ + pip install 'ddpm_guided_diffusion' -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html --no-index && \ + pip uninstall shotdetect_scenedetect_lgss -y && \ + pip install 'shotdetect_scenedetect_lgss' -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html --no-index && \ + pip uninstall MinDAEC -y && \ + pip install https://modelscope.oss-cn-beijing.aliyuncs.com/releases/dependencies/MinDAEC-0.0.2-py3-none-any.whl && \ + pip cache purge; \ +else \ + pip install --no-cache-dir -r /var/modelscope/framework.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + pip cache purge; \ +fi + +ARG CUR_TIME={cur_time} +RUN echo $CUR_TIME + +RUN curl -fsSL https://ollama.com/install.sh | sh && \ + pip install --no-cache-dir -U funasr scikit-learn && \ + pip install --no-cache-dir -U qwen_vl_utils qwen_omni_utils librosa timm transformers accelerate peft trl safetensors && \ + cd /tmp && GIT_LFS_SKIP_SMUDGE=1 git clone -b {swift_branch} --single-branch https://github.com/modelscope/ms-swift.git && \ + cd ms-swift && pip install .[llm] && \ + pip install .[eval] && pip install evalscope -U --no-dependencies && pip install ms-agent -U --no-dependencies && \ + cd / && rm -fr /tmp/ms-swift && pip cache purge; \ + cd /tmp && GIT_LFS_SKIP_SMUDGE=1 git clone -b {modelscope_branch} --single-branch https://github.com/modelscope/modelscope.git && \ + cd modelscope && pip install . -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \ + cd / && rm -fr /tmp/modelscope && pip cache purge; \ + pip install --no-cache-dir transformers diffusers timm>=0.9.0 && pip cache purge; \ + pip install --no-cache-dir omegaconf==2.3.0 && pip cache purge; + + +RUN if [ "$INSTALL_MS_DEPS" = "True" ]; then \ + pip install --no-cache-dir huggingface-hub transformers peft -U; \ +fi; + +ENV SETUPTOOLS_USE_DISTUTILS=stdlib +ENV VLLM_USE_MODELSCOPE=True +ENV LMDEPLOY_USE_MODELSCOPE=True +ENV MODELSCOPE_CACHE=/mnt/workspace/.cache/modelscope/hub + +# Show install results +RUN pip list + +SHELL ["/bin/bash", "-c"] diff --git a/docker/build_image.py b/docker/build_image.py index 07c616f4..916bb5c2 100644 --- a/docker/build_image.py +++ b/docker/build_image.py @@ -415,6 +415,38 @@ RUN pip install --no-cache-dir -U icecream soundfile pybind11 py-spy return os.system(f'docker push {image_tag2}') +class AscendSwiftImageBuilder(SwiftImageBuilder): + def init_args(self, args) -> Any: + if not args.base_image: + # other vision search for: https://hub.docker.com/r/ascendai/cann/tags + args.base_image = "swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.3.rc1-a3-ubuntu22.04-py3.11" + return super().init_args(args) + + def generate_dockerfile(self) -> str: + extra_content = """ +RUN pip install --no-cache-dir -U icecream soundfile pybind11 py-spy +""" + with open('docker/Dockerfile.ascend', 'r') as f: + content = f.read() + content = content.replace('{base_image}', self.args.base_image) + content = content.replace('{extra_content}', extra_content) + content = content.replace('{cur_time}', formatted_time) + content = content.replace('{install_ms_deps}', 'False') + content = content.replace('{modelscope_branch}', + self.args.modelscope_branch) + content = content.replace('{swift_branch}', self.args.swift_branch) + return content + + def image(self) -> str: + return ( + f'{docker_registry}:{self.args.base_image.split(":")[-1]}-torch2.7.1' + f'-{self.args.modelscope_version}-ascend-swift-test' + ) + + def push(self): + return 0 + + parser = argparse.ArgumentParser() parser.add_argument('--base_image', type=str, default=None) parser.add_argument('--image_type', type=str) @@ -448,6 +480,8 @@ elif args.image_type.lower() == 'llm': builder_cls = LLMImageBuilder elif args.image_type.lower() == 'swift': builder_cls = SwiftImageBuilder +elif args.image_type.lower() == 'ascend_swift': + builder_cls = AscendSwiftImageBuilder else: raise ValueError(f'Unsupported image_type: {args.image_type}')