poweruser
9d@n2N3Vt5Z6
sudo groupadd docker
sudo usermod -aG docker ${USER}
sudo systemctl restart docker
dd if=/dev/zero of=test.bin bs=1k count=1000
export http_proxy=http://192.168.100.200:3128
export https_proxy=http://192.168.100.200:3128
apt update && apt install -y sshfs
yum clean all && yum update
yum install epel-release
yum install fuse-sshfs
sshfs -o allow_root,default_permissions [email protected]:/data /mnt/fileserver/
corex@2018
sshfs -o allow_root,default_permissions [email protected]:/mnt/nas/psr /mnt/nas
ln -s /mnt/nas/Home/lei.zhang/remote_src remote_src
mount -t nfs 10.113.1.9:/mnt/nas/psr /mnt/nas
mount -t nfs 10.113.3.3:/data /mnt/nas
docker login -u lei.zhang1 10.150.9.98:80
# ping 工具安装
apt-get install inetutils-ping
# git 忽略权限配置
git config --global core.filemode false
git config core.filemode false
# git 配置默认编辑器vim
git config --global core.editor "vim"
# 查看所有分支
git branch -a
# checkout 分支
git checkout -b xxxx（本地分支名称） yyyy(上条命令查找到的远程分支的名称)
# 同步所有远程分支
git branch -r | grep -v '\->' | while read remote; do git branch --track "${remote#origin/}" "$remote"; done
git pull --all
# 大家在提交代码时，在机器上把git config global设置都给清除了。
git config --global user.name "lei.zhang1"
git config --global user.email "[email protected]"
# 统一在自己项目目录使用--local参数：
git config --local user.name "lei.zhang1"
git config --local user.email "[email protected]"
git config --global user.name "chenugray"
git config --global user.email "[email protected]"
# docker
docker run -it --name igie_zhanglei_0513 --privileged --cap-add=ALL -v /dev:/dev -v /root:/root -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang 10.150.9.98:80/sw_test/apps:2.3.0.20221115.265-centos-7.8.2003-x86_64-10.2-python3.7_mr /bin/bash
docker run --name "container_bi_20220422" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang container_bi_20220422-1130 /bin/bash
docker run --name "tvm0.9dev_leizhang" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang 2912baf2ad1 /bin/bash
docker run --name "bi_img_leizhang_20220210" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang -v /mnt/nas/home/lei.zhang/:/mnt/nas/home/lei.zhang/ 10.150.9.98:80/performance/bi_img:20220210 /bin/bash
docker run --name "bi_img_leizhang_c" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang container_bi_20220422-1130:1.0 /bin/bash
sudo docker run --name "tvm0.9dev_leizhang_21.10" --shm-size=16g --ulimit --ulimit stack=67108864 -it -v /home/lei.zhang/:/home/lei.zhang/ --privileged nvcr.io/nvidia/tensorrt:21.10-py3
docker run --name "mr_0810_latest" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang 10.150.9.98:80/sw_test/apps:mr_alpha_20220809-centos-7.8.2003-x86_64-10.2-python3.7_mr /bin/bash
## docker 拉起pytorch gpu镜像
sudo nvidia-docker run --name "zhanglei.cuda11.7_llama" --ulimit memlock=-1 --ulimit stack=67108864 -it -v /home/lei.zhang/:/home/lei.zhang/ -v /mnt/:/mnt/ -v /data0:/data0 -v /data:/data --privileged nvcr.io/nvidia/cuda:11.7.1-cudnn8-devel-ubuntu20.04
sudo nvidia-docker run --gpus all --shm-size=16g --ulimit memlock=-1 --ulimit stack=67108864 -it -v /home/lei.zhang/:/home/lei.zhang/ --privileged nvcr.io/nvidia/tensorrt:22.02-py3
docker run --name "lei.zhang230215" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang -v /mnt/nas/home/lei.zhang/remote_src/:/home/lei.zhang/src_code/remote_src/ 10.150.9.98:80/sw_test/apps@sha256:381596743f46e4b13b41c7b891f658d985b81d664758d9cc698f5905be93c2cc /bin/bash
sudo docker exec -it lei.igie_zhanglei_0513 /bin/bash
sudo docker exec -it distracted_pasteur /bin/bash
bi_img_leizhang_20220217
sudo docker exec -it bi_img_leizhang_20220217 /bin/bash
sudo docker exec -it container_leizhang_20220606-1756 /bin/bash
container_bi_20220422-1130
sudo docker exec -it container_bi_20220422-1130 /bin/bash
sudo docker exec -it zhanglei_tensorrt_22.02 /bin/bash
sudo docker exec -it mr_0810_latest /bin/bash
reverent_lamarr
# 将运行的镜像保存为可加载镜像，以便后续拉起
sudo docker commit 14478deb892d tvm_gpu_test_images
sudo docker run --name "tf1_py3_lei_zhang" --gpus all --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 -it --rm -v /home/lei.zhang/:/home/lei.zhang/ -v /mnt/nas/:/mnt/nas/ --privileged nvcr.io/nvidia/tensorflow:22.07-tf1-py3
sudo -s
# adduser user.name --force-badname
useradd -m -d /home/lei.zhang -s /bin/bash lei.zhang
usermod -a -G sudo fanwu.han
# 格式化时间
date "+%Y_%m_%d %H:%M:%S" # 2022_03_11 10:00:13
date "+%Y_%m_%d_%H_%M" # 2022_03_11_10_01
MR的1.0.5固件可以从以下位置获取：
/mnt/nas/home/yuanpei.wang/tools/MR/FW/1.0.5/mr50_75w_v1.0.5
/mnt/nas/home/yuanpei.wang/tools/MR/FW/1.0.5/mr100_150w_v1.0.5
使用方法如下：
./ixfw-tool -f mr50_75w_v1.0.5.hex
./ixfw-tool -f mr100_150w_v1.0.5.hex
10.150.9.98:80/sw_test/apps@sha256:043ac7e87013bff65e6f17f18e7c9756282970df32da6983543f091b34f0dedb
docker pull 10.150.9.98:80/sw_test/apps:latest-centos-7.8.2003-x86_64-10.2-python3.6_mr
sudo docker run --name "lei.zhang_0308" -it --privileged --pid=host --cap-add=ALL -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/lei.zhang:/home/lei.zhang 10.150.9.98:80/sw_test/apps@sha256:043ac7e87013bff65e6f17f18e7c9756282970df32da6983543f091b34f0dedb /bin/bash
sudo docker run --name "gpu_pytorch_v6_zhanglei" --gpus all --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 -it -v /home/lei.zhang/:/home/lei.zhang/ --privileged cmcc:conformer
sudo docker exec -it lei.igie_zhanglei_0513 /bin/bash
ssh [email protected]@[email protected] -p 2222
ssh [email protected]@[email protected] -p 2222
ssh [email protected]@[email protected] -p 2222
10.113.3.3
root
corex@2018
source /opt/rh/devtoolset-9/enable$$ l_shift = s > 0 ? s : 0\ r_shift = s > 0 ? 0 : -s\
z = \frac{x * 2^{l_shift} * y + 2^{r_shift + q - 1}}{2^{r_shift+q}} = x * y * 2^{l_shift - r_shift - q} + 2 ^ {-1} \ for\ any \ s \ always \ has \ (l_shift - r_shift) = s \ z = x * y * 2 ^ {s - q} + 0.5 \ scale = y * 2 ^ {s - q} \ z = round(x * scale + 0.5) $$
该模型是transformer结构的encoder+decoder模型进行translation任务，具体来源如下
fairseq translation en2fr example: https://github.com/facebookresearch/fairseq/tree/main/examples/translation
当前推荐使用64batch和128batch进行推理，使用其他batch可能会产生性能退化现象, 这里给出参考8/16/32/64/128 batch 参考性能。
| Column 1 | Column 2 |
|---|---|
| Row 1, Column 1 | Row 1, Column 2 |
| Row 2, Column 1 | Row 2, Column 2 |
|| batch_size || performance || accuracy(BLUE4) || | ---- | ------------ | ----------- | | 8 | 9.32 sentences/s | 42.34 | | 16 | 15.71 sentences/s | 42.34 | | 32 | 24.31 sentences/s | 42.34 | | 64 |265.17 sentences/s | 42.34 | | 128 |301.87 sentences/s | 42.34 |