ray使用手册

1. 安装conda：wget https://repo.anaconda.com/archive/Anaconda3-2021.05-Linux-x86_64.sh

2. bash Anaconda3-2021.05-Linux-x86_64.sh

export PATH=/home/zjy/anaconda3/bin:$PATH
3. source ~/.bashrc

4. conda create -n ray python==3.6.6

5. conda activate ray

6. pip install ray==1.5.1 -i https://pypi.tuna.tsinghua.edu.cn/simple
   conda insatll ray ==1.6.0

7. PS机器学习tutorial：pip install torch torchvision filelock -i https://pypi.tuna.tsinghua.edu.cn/simple

   https://docs.ray.io/en/master/auto_examples/plot_parameter_server.html

8. 示例code：test.py
   import ray
   import time
   ray.init() #local mode

@ray.remote
#@ray.remote(num_gpus=0.1)
def f1(x):
#time.sleep(1)
a = x * x
return a

def f2(x):
#time.sleep(1)
a = x * x
return a

futures = [f.remote(i) for i in range(4)]
print(ray.get(futures))

9. 运行code：python test.py

-------------------暂时不需要看下面内容----------------------

10.build ray by source
cd python/ray/new_dashboard/client && npm install && npm ci && npm run build): ‘/home/yangkun/anaconda3/envs/ray/lib/python3.6/site-packages/ray/new_dashboard/client/build’

https://docs.ray.io/en/master/development.html

curl -sL https://deb.nodesource.com/setup_10.x | sudo -E bash -
apt -y install nodejs make gcc g++

/tmp/ray/session_latest/logs/raylet.err 可以查看日志
https://docs.ray.io/en/master/debugging.html

python -m ipykernel install --user --name ray --display-name “ray”

pip install -e . --verbose --user
pip3 install -e . --verbose --user