用户需求：

按用户需求进行组网部署，提供硬件驱动、正版操作系统、网络配置，DeepSeek671B部署、模型基本安装部署并测试验证。

操作流程：

参考

华为文档：https://support.huawei.com/enterprise/zh/doc/EDOC1100494820/a045793a

服务器部署：

部署操作系统

docker环境

Ubuntu 22.04
curl -fsSL https://repo.huaweicloud.com/docker-ce/linux/debian/gpg | sudo apt-key add -
apt-get install -y docker-ce

CTyunOS 22.06、openEuler 22.03 LTS、UOS20 1050e、CULinux 3.0、Kylin V10 SP2、Kylin V10 SP3、BC-Linux 21.10
yum install -y docker

昇腾NPU固件、驱动

检查显卡是否在线

lspci | grep d500

确认系统版本和内核版本与下表对应

uname -m && cat /etc/*release
uname -r

关闭内核自动更新

yum install yum-plugin-versionlock
yum versionlock add kernel-$(uname -r)

执行如下命令检查：

make -v
rpm -qa | grep dkms
rpm -qa | grep gcc
rpm -qa | grep kernel-devel-$(uname -r)

若回显相关软件包版本信息，表示已安装；
若未安装请执行yum install -y make dkms gcc kernel-devel-$(uname -r)命令安装依赖。

创建安装驱动用户

groupadd HwHiAiUser
useradd -g HwHiAiUser -d /home/HwHiAiUser -m HwHiAiUser -s /bin/bash

增加执行权限

chmod +x Ascend-hdk-xxx-npu-driver_<version>_linux-aarch64.run
chmod +x Ascend-hdk-xxx-npu-firmware_<version>.run

首次安装先安装驱动再安装固件

安装驱动

./Ascend-hdk-xxx-npu-driver_<version>_linux-aarch64.run --full --install-for-all

安装固件

./Ascend-hdk-xxx-npu-firmware_<version>.run --full

若系统出现如下关键回显信息，表示固件安装成功。

Firmware package installed successfully! Reboot now or after driver installation for the installation/upgrade to

验证状态

npu-smi info

下载大模型

方法一：使用python

pip install huggingface_hub -i https://pypi.tuna.tsinghua.edu.cn/simple

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="deepseek-ai/DeepSeek-R1-0528",
    local_dir="/data/DeepSeek-R1",
    local_dir_use_symlinks=False,
    cache_dir="/data/cache"
)

from huggingface_hub import snapshot_download
import os

# 设置国内镜像（可选，如果下载速度慢）
os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com'

snapshot_download(
    repo_id="deepseek-ai/DeepSeek-R1-0528",
    local_dir="/data/DeepSeek-R1",
    local_dir_use_symlinks=False,
    cache_dir="/data/cache"
)

方法二：使用git

yum install git-lfs -y
git lfs install 
git clone https://huggingface.co/deepseek-ai/DeepSeek-R1 ./DeepSeek-R1-671B

#或者镜像源
git clone https://hf-mirror.com/deepseek-ai/DeepSeek-R1 ./DeepSeek-R1-671B

方法三：国内仓库

pip install modelscope -i https://pypi.tuna.tsinghua.edu.cn/simple

modelscope download --model deepseek-ai/DeepSeek-R1-0528 --local_dir ./DeepSeek-R1-671B

或者python下载
from modelscope import snapshot_download

model_dir = snapshot_download(
    'deepseek-ai/DeepSeek-R1-0528',
    local_dir="/data/DeepSeek-R1",
    local_dir_use_symlinks=False,
    cache_dir="/data/cache"
)

或者git下载
git lfs install
git clone https://www.modelscope.cn/deepseek-ai/DeepSeek-R1-0528.git

容器环境：

下载镜像

下载镜像：mindie_1.0.RC2_aarch64.tar，此镜像包含CANN(Toolkit、kernels、NNAL)、MindIE、ChatGLM3-6B模型权重和推理脚本，可用于大模型推理。

链接：https://gpt-3.obs.cn-north-4.myhuaweicloud.com/ChatGLM3/mindie_1.0.RC2_aarch64.tar

官方sha256值：4e00c9eec37a5f266331358940409d1ba6e122cd80d1d4851ce3ac3b415be08b

加载镜像

docker load < mindie_1.0.RC2_aarch64.tar

这样用于验证测试，生产环境镜像根据系统、显卡、驱动版本选择

步骤

参考仓库地址：https://www.hiascend.com/developer/ascendhub/

启动容器

docker run -itd -u root --net=host --ipc=host --privileged --shm-size=500g --name=ds_r1 \
--device=/dev/davinci0 \
--device=/dev/davinci1 \
--device=/dev/davinci2 \
--device=/dev/davinci3 \
--device=/dev/davinci4 \
--device=/dev/davinci5 \
--device=/dev/davinci6 \
--device=/dev/davinci7 \
--device=/dev/davinci_manager \
--device=/dev/devmm_svm \
--device=/dev/hisi_hdc \
-v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
-v /usr/local/Ascend/add-ons/:/usr/local/Ascend/add-ons/ \
-v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \
-v /usr/local/sbin/:/usr/local/sbin/ \
-v /var/log/npu/conf/slog/slog.conf:/var/log/npu/conf/slog/slog.conf \
-v /var/log/npu/slog/:/var/log/npu/slog \
-v /var/log/npu/profiling/:/var/log/npu/profiling \
-v /var/log/npu/dump/:/var/log/npu/dump \
-v /var/log/npu/:/usr/slog \
-v /data/:/data/ \
mindie镜像id \
/bin/bash

mindie:2.1.RC2-800I-A2-py311-openeuler24.03-lts

进入设置变量

docker exec -it ds_r1 bash

source /usr/local/Ascend/mindie/set_env.sh
source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh
source /usr/local/Ascend/atb-models/set_env.sh

# 根据自己的服务器设置ranktable_file_path，machine_IP， master_IP
export RANK_TABLE_FILE="/usr/local/Ascend/mindie/latest/mindie-service/ranktable.json"
export MIES_CONTAINER_IP='x.x.x.x'
export MASTER_IP='x.x.x.x'

export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_HOST_SOCKET_PORT_RANGE=60000-60050
export HCCL_CONNECT_TIMEOUT=7200
export MASTER_PORT=7008
export MINDIE_LOG_TO_STDOUT=1

export ATB_LLM_ENABLE_AUTO_TRANSPOSE=0

vim /usr/local/Ascend/atb-models/atb_llm/layers/linear/linear.py +112

out = torch_npu.npu_quant_matmul(input_tensor_quant, self._weight.transpose(-2, -1), self._weight_scale, pertoken_scale=pertoken_scale, bias=None, output_dtype=torch.bfloat16)