docker等基础工具使用

普通docker启动常用指令：

sudo docker run --name XXX --shm-size="200g" --privileged=true --device /dev/biren -it -d -v /xx:/xx --net=host xxximage /bin/bash

nv 环境docker启动常用指令

docker run -it --gpu=all --ipc=host --network=host --name xxx -v /xx:/xx xximage /bin/bash

slurm 启动

ENROOT_ALLOW_SUPERUSER=yes \

ENROOT_ROOTFS_WRITABLE=yes \

PYTORCH_VERSION=1 \

MELLANOX_VISIBLE_DEVICES=all \

MELLANOX_MOUNT_DRIVER=1 \

SRUN_CHROOT=/PATH \

srun -w server$1\

        --container-env=ENROOT_ALLOW_SUPERUSER, ENROOT_ROOTFS_WRITABLE, PYTORCH_VERSION,MELLANOX_VISIBLE_DEVICES,MELLANOX_MOUNT_DRIVER \

        --container-image=xxx \

        --container-mounts=/xxx:/xxx \

        --container-name='xxx' \

        --pty bash

GIT 合并一个MR所有commit至一个

GITHUB:  git fetch origin pull/188/head -> git merge FETCH_HEAD

GITLAB:  git fetch origin refs/merge-requests/22/head:pr22 -> git merge pr22

git 消除文件权限带来的改动显示： git config core.filemode false

DEBUG

import pdb; pdb.set_trace()

from remote_pdb import set_trace; set_trace()

gdb attach (os.getpid())

def info_log:

        if torch.distribued.get_rank() == 0:

                print(f"{} -- {} -- {}")