环境信息：
npu-smi info
+--------------------------------------------------------------------------------------------------------+
| npu-smi 25.3.rc1.b030 Version: 25.3.rc1.b030 |
+-------------------------------+-----------------+------------------------------------------------------+
| NPU Name | Health | Power(W) Temp(C) Hugepages-Usage(page) |
| Chip Device | Bus-Id | AICore(%) Memory-Usage(MB) |
+===============================+=================+======================================================+
| 2 310P3 | OK | NA 64 0 / 0 |
| 0 0 | 0000:01:00.0 | 0 1556 / 44280 |
+-------------------------------+-----------------+------------------------------------------------------+
| 2 310P3 | OK | NA 63 0 / 0 |
| 1 1 | 0000:01:00.0 | 0 1413 / 43693 |
+===============================+=================+======================================================+
| 5 310P3 | OK | NA 73 0 / 0 |
| 0 2 | 0000:81:00.0 | 0 1483 / 44280 |
+-------------------------------+-----------------+------------------------------------------------------+
| 5 310P3 | OK | NA 72 0 / 0 |
| 1 3 | 0000:81:00.0 | 0 1481 / 43693 |
+===============================+=================+======================================================+
| 6 310P3 | OK | NA 74 0 / 0 |
| 0 4 | 0000:82:00.0 | 0 1466 / 44280 |
+-------------------------------+-----------------+------------------------------------------------------+
| 6 310P3 | OK | NA 71 0 / 0 |
| 1 5 | 0000:82:00.0 | 0 1500 / 43693 |
+===============================+=================+======================================================+
+-------------------------------+-----------------+------------------------------------------------------+
| NPU Chip | Process id | Process name | Process memory(MB) |
+===============================+=================+======================================================+
lscpu
架构： aarch64
CPU 运行模式： 64-bit
字节序： Little Endian
CPU: 96
在线 CPU 列表： 0-95
厂商 ID： HiSilicon
BIOS Vendor ID: HiSilicon
型号名称： Kunpeng-920
BIOS Model name: HUAWEI Kunpeng 920 5250
型号： 0
每个核的线程数： 1
每个座的核数： 48
座： 2
步进： 0x1
Frequency boost: disabled
CPU 最大 MHz： 2600.0000
CPU 最小 MHz： 200.0000
BogoMIPS： 200.00
标记： fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma dcpop asimddp asimdfhm
Caches (sum of all):
L1d: 6 MiB (96 instances)
L1i: 6 MiB (96 instances)
L2: 48 MiB (96 instances)
L3: 96 MiB (4 instances)
NUMA:
NUMA 节点： 4
NUMA 节点0 CPU： 0-23
NUMA 节点1 CPU： 24-47
NUMA 节点2 CPU： 48-71
NUMA 节点3 CPU： 72-95
Vulnerabilities:
Gather data sampling: Not affected
Itlb multihit: Not affected
L1tf: Not affected
Mds: Not affected
Meltdown: Not affected
Mmio stale data: Not affected
Reg file data sampling: Not affected
Retbleed: Not affected
Spec rstack overflow: Not affected
Spec store bypass: Not affected
Spectre v1: Mitigation; __user pointer sanitization
Spectre v2: Not affected
Srbds: Not affected
Tsx async abort: Not affected
CANN版本：8.2.RC1
用到的镜像：mindie:2.1.RC1.B152-300I-Duo-py3.11-openeuler24.03-lts-aarch64
##输入npu-smi info，若有卡的信息则表示有NPU卡
npu-smi info##通过能否curl通百度来确定
curl www.baidu.com
##若不能curl通则需要配置代理，cntlm代理的配置参考https://3ms.huawei.com/km/blogs/details/21430554
##例如：
CCW_HOST_IP=90.254.50.56
export http_proxy="http://${CCW_HOST_IP}:3128"
export https_proxy=${http_proxy}
export ftp_proxy=${http_proxy}
export GIT_SSL_NO_VERIFY=true
##其中的CCW_HOST_IP需要更换成VPN的地址
##输入docker -h可以检查
docker -h
##若没有docker则需要安装
yum install docker##以A3的openeuler镜像为例
docker pull mindie:2.1.RC1.B152-300I-Duo-py3.11-openeuler24.03-lts-aarch64
##通过docker images来查看已经pull的镜像
####
[root@localhost ~]# docker images
export IMAGE=mindie:dev-2.2.RC1.B110-300I-Duo-py311-ubuntu22.04-aarch64 && docker run --privileged -u root --name mindie-ljh-1014 --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/dcmi:/usr/local/dcmi -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info -v /etc/ascend_install.info:/etc/ascend_install.info -v /root/.cache:/root/.cache -p 18050:8080 -it $IMAGE bashyum install -y git
yum install -y patch
yum install -y unzip
yum install -y mesa-libGL
yum install gcc-c++##自己创建一个文件夹 比如/home/ljh，并cd /home/ljh
git clone https://gitee.com/ascend/ModelZoo-PyTorch.git
cd ModelZoo-PyTorch/ACL_PyTorch/built-in/cv/ViTDet_for_Pytorch
###若遇到git访问权限问题可以通过如下命令解决
git config --global --unset http.proxy
git config --global --unset https.proxygit clone https://github.com/open-mmlab/mmdetection.git
git clone https://github.com/open-mmlab/mmengine.git
cd mmdetection
git reset --hard cfd5d3a985b0249de009b67d04f37263e11cdf3d
pip3 install -r requirements.txt
cd ../mmengine
git reset --hard 390ba2fbb272816adfd2883642326d0fd0ca6049
pip3 install -r requirements.txt
cd ..
##从源码安装mmcv
git clone https://github.com/open-mmlab/mmcv.git
cd mmcv
git checkout v2.2.0
pip install -r requirements.txt
MMCV_WITH_OPS=1 pip install -e .
若遇到AttributeError: type object 'Callable' has no attribute '_abc_registry'报错则可通过下面命令解决：
rm -rf /usr/local/lib/python3.11/site-packages/typing.py
rm -rf /usr/local/lib/python3.11/site-packages/typing-*
# 确保用系统对应的 python3.11 来执行
python3.11 -m pip uninstall -y typing
python3.11 -m pip uninstall -y typing_extensions
若遇到Using cached fastmcp-2.9.0-py3-none-any.whl.metadata (17 kB)安装特别慢或aim<=3.17.5安装失败等错误可以先注释以下代码：
然后再通过pip install 单独安装
vim requirements/tests.txt
#aim<=3.17.5;sys_platform!='win32'
bitsandbytes
clearml
coverage
dadaptation
dvclive
lion-pytorch
lmdb
#mlflow
parameterized
pydantic==1.10.9
pytest
transformers
cp mmengine.patch mmengine/mmengine/
cp mmdet.patch mmdetection/
cp infer.py mmdetection/cd mmengine/mmengine/
patch -p2 < mmengine.patch
pip install -v -e ..
cd ../../mmdetection
patch -p1 < mmdet.patchpip install -v -e .. 时遇到 ImportError: cannot import name 'Logger' from partially initialized module 'logging' (most likely due to a circular import) 错误的解决办法cd ViTDet_for_Pytorch/mmengine
pip install -v -e .
###下载权重文件并放到mmdetection/ckpt下
cd mmdetection
mkdir ckpt
cd ckpt
wget --no-check-certificate https://download.openmmlab.com/mmdetection/v3.0/vitdet/vitdet_mask-rcnn_vit-b-mae_lsj-100e/vitdet_mask-rcnn_vit-b-mae_lsj-100e_20230328_153519-e15fe294.pth
###下载数据集
cd ..
mkdir data
cd data
mkdir coco
cd coco
# 下载验证集图片
wget http://images.cocodataset.org/zips/val2017.zip
unzip val2017.zip
rm val2017.zip
# 下载标注
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
unzip annotations_trainval2017.zip
rm annotations_trainval2017.zip
11模型推理
在310P的环境上推理前需修改以下代码：
由于当前numpy的版本较高，直接推理会报错，需要降低numpy的版本：
pip uninstall numpy
pip install numpy==1.24.4
pip install future tensorboardsource /usr/local/Ascend/ascend-toolkit/set_env.sh
python infer.py --cfg projects/ViTDet/configs/vitdet_mask-rcnn_vit-b-mae_lsj-100e.py --ckpt "ckpt/vitdet_mask-rcnn_vit-b-mae_lsj-100e_20230328_153519-e15fe294.pth" --warm_up_times 1
##首先卸载torch等
pip uninstall torch torchvision torch_npu
##然后重新安装torch 等
pip install torch==2.4.0 torch_npu==2.4.0post4 torchvision==0.19.0 torchaudio==2.4.0
##参考：https://www.hiascend.com/document/detail/zh/Pytorch/710/releasenote/releasenote_0003.htmlCCW_HOST_IP=141.4.102.160 # 141.4.124.251
export http_proxy="http://${CCW_HOST_IP}:3128"
export https_proxy=${http_proxy}
export ftp_proxy=${http_proxy}
export GIT_SSL_NO_VERIFY=true参考教程：gitee.com
IMAGES_ID=$1
NAME=$2
if [ $# -ne 2 ]; then
echo "error: need one argument describing your container name."
exit 1
fi
docker run --name ${NAME} -it -d --net=host --shm-size=500g \
--privileged=true \
-w /home \
--device=/dev/davinci_manager \
--device=/dev/hisi_hdc \
--device=/dev/devmm_svm \
--entrypoint=bash \
-v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
-v /usr/local/dcmi:/usr/local/dcmi \
-v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
-v /usr/local/sbin:/usr/local/sbin \
-v /home:/home \
-v /tmp:/tmp \
-v /usr/share/zoneinfo/Asia/Shanghai:/etc/localtime \
-e http_proxy=$http_proxy \
-e https_proxy=$https_proxy \
${IMAGES_ID}
bash start-docker.sh 5718854ec902 dev2_rc1_b010
5718854ec902 为IMAGES_ID，dev2_rc1_b010为NAME
pip3 install --trusted-host pypi.org --trusted-host files.pythonhosted.org -r requirements.txt
或者 pip3 install *** -i https://pypi.tuna.tsinghua.edu.cn/simple
报错解决办法：降级numpy包
mkdir -p ~/.pip
vim ~/.pip/pip.conf
[global]
index-url = https://pypi.tuna.tsinghua.edu.cn/simple
timeout = 6000
[install]
trusted-host = pypi.tuna.tsinghua.edu.cn910B上将pse设置为None跑出来的结果：（相比较上面下降30%）
