个人环境常用命令

Anaconda

wget https://repo.anaconda.com/archive/Anaconda3-5.3.0-Linux-x86_64.sh
chmod +x Anaconda3-5.3.0-Linux-x86_64.sh
./Anaconda3-5.3.0-Linux-x86_64.sh
source ~/.bashrc

清华源

conda config --show channels
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/r
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/pro
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/msys2

创建 激活 删除环境

conda create -n xxx python=3.10
conda activate xxx
conda remove -n xxx --all

docker

修改 docker image 路径

sudo vim /etc/docker/daemon.json

添加 data-root

{
	"data-root": "/logging/xxxxx"
}

配置生效

sudo systemctl daemon-reload
sudo systemctl restart docker
sudo docker info

或者

sudo vim /usr/lib/systemd/system/pouch.service
在 ExecStart 后面添加 -g path

Pytorch

官网下载：https://pytorch.org/
根据cuda版本修改后缀 cu113/117/121

pip3 config set global.index-url https://mirrors.aliyun.com/pypi/simple
pip install torch torchvision torchaudio -i https://download.pytorch.org/whl/cu117

测试

import torch
print(torch.__version__)
print(torch.cuda.is_available())
print(torch.cuda.device_count())

LLM

Huggingface

export HF_ENDPOINT=https://hf-mirror.com
huggingface-cli download --resume-download --local-dir-use-symlinks False mistralai/Mistral-7B-v0.1 --local-dir Mistral-7B-v0.1

分布式训练

export NCCL_IB_DISABLE=1
export NCCL_SOCKET_IFNAME=$(ip route | awk ‘/default/ { print $5 }’)
export CUDA_VISIBLE_DEVICES=0,1,2

vllm

pip install --editable ./ --no-build-isolation -i https://mirrors.aliyun.com/pypi/simple
CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server --model /local_workdir/save_model/xxxxx --trust-remote-code --tensor-parallel-size 2 --port 18500 --host 0.0.0.0 --gpu-memory-utilization 0.9

请求

curl http://xx.xx.xx.xx:18500/v1/completions -H "Content-Type: application/json" -d '{
        "model": "/local_workdir/save_model/xxxx",
        "prompt": "<|im_start|>user\n你是谁 <|im_end|>\n <|im_start|>assistant\n",
        "max_tokens": 500,
        "temperature": 0.3,
        "frequency_penalty" : 1,
        "stop" : ["<|im_end|>","<|im_start|>"]
    }'

apex

git clone https://github.com/NVIDIA/apex
cd apex
git checkout 2386a912164b0c5cfcd8be7a2b890fbac5607c82
pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./

或者

git checkout 22.04-dev
pip install -r requirements.txt
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

磁盘空间占用

sudo df ./
sudo du -h --max-depth=1

GCC 8.3.0

安装GMP：

wget http://mirrors.nju.edu.cn/gnu/gmp/gmp-6.1.2.tar.xz
tar -Jxf gmp-6.1.2.tar.xz
rm -rf gmp-6.1.2.tar.xz
mkdir /local_workdir/env/gmp
cd gmp-6.1.2
./configure --prefix=/local_workdir/env/gmp
make
make install

安装MPFR：

wget http://mirrors.nju.edu.cn/gnu/mpfr/mpfr-4.0.2.tar.xz
tar -Jxf mpfr-4.0.2.tar.xz
rm -rf mpfr-4.0.2.tar.xz
mkdir /local_workdir/env/mpfr
cd mpfr-4.0.2
./configure --prefix=/local_workdir/env/mpfr --with-gmp=/local_workdir/env/gmp
make
make install

安装mpc：

wget http://mirrors.nju.edu.cn/gnu/mpc/mpc-1.1.0.tar.gz
tar -xzvf mpc-1.1.0.tar.gz
rm -rf mpc-1.1.0.tar.gz
mkdir //local_workdir/env/mpc
cd mpc-1.1.0
./configure --prefix=/local_workdir/env/mpc --with-gmp=/local_workdir/env/gmp --with-mpfr=/local_workdir/env/mpfr
make -j4&& make install

安装gcc：

wget http://mirrors.nju.edu.cn/gnu/gcc/gcc-8.3.0/gcc-8.3.0.tar.gz
tar -xzvf gcc-8.3.0.tar.gz
rm -rf gcc-8.3.0.tar.gz
mkdir /local_workdir/env/gcc
./configure --prefix=/local_workdir/env/gcc --disable-multilib --with-gmp=/local_workdir/env/gmp --with-mpfr=/local_workdir/env/mpfr --with-mpc=/local_workdir/env/mpc
make
make install

添加环境变量

nano ～/.bashrc
export LD_LIBRARY_PATH=/local_workdir/env/gmp/lib:/local_workdir/env/mpfr/lib:/local_workdir/env/mpc/lib$LD_LIBRARY_PATH
export PATH=/local_workdir/env/gcc/bin:$PATH
export MANPATH=/local_workdir/env/gcc/share/man:$MANPATH
source ～/.bashrc

maven 3.6.3

https://archive.apache.org/dist/maven/maven-3/3.6.3/binaries/ 下载安装包 bin.tar.gz
添加环境变量：

nano ~/.bashrc
export MAVEN_HOME=/xxxxx/runenv/apache-maven-3.6.3
export PATH=$PATH:$MAVEN_HOME/bin
source ~/.bashrc

测试

mvn -v