Installing Dependencies

All data in WebCal is retained after shutdown, including installed dependencies; there is no need to reinstall them upon restart. See Retention Rules. If you wish to reuse your dependency environment when renting other hosts, you can save all data on the system disk as an image or migrate the instance. Please refer to Saving/Loading Images and Migrating Instances.

We recommend using domestic mirrors to speed up the download of dependency packages. However, if a particular mirror is slow during peak network traffic, you can switch to another mirror to improve download speed. See Software Mirrors.

Installing Other Versions of Python

All images include Miniconda.

# 比如构建一个虚拟环境名为：my-env，Python版本为3.7
conda create -n my-env python=3.7

# 更新bashrc中的环境变量
conda init bash && source /root/.bashrc

# 切换到创建的虚拟环境：my-env
conda activate my-env

Installing Other Frameworks

# PyTorch：从https://pytorch.org/get-started/previous-versions/找到合适的版本。比如：
pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1

# TensorFlow: 从https://www.tensorflow.org/install/pip找到对应版本链接。比如：
pip install https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-2.6.0-cp38-cp38-manylinux2010_x86_64.whl

Installing Python Dependencies

# 使用pip，举例
pip install opencv-python scipy numpy Pillow

# 使用conda，举例
conda install numpy

# 如果不确定包名，通过搜索来查找
pip search xxxx
conda search xxxx

如果在用pip时不知道某个依赖有什么版本号，那么以下trick可以查看：随便写一个版本号，pip会报错出所有可安装的版本号
$ pip install xxx=9.9

Installing System Dependencies

# 以安装zip为例
apt-get update    # 只需要执行一次，不用每次都执行
apt-get install -y zip

# 如果不知道包名
apt-get update    # 只需要执行一次，不用每次都执行
apt-cache search xxxxx

For instructions on installing CUDA and other software, please refer to: Documentation