origin
The reason is that the training server is restricted from accessing the Internet during the security period. Moreover, the training server that wants to connect to the A100 graphics card must be connected through the bastion host (that is, the target server cannot be connected to the Internet, nor is it in the same network segment as the local machine)
Installation tools
First, use conda env listto view all environments. Make sure that the environment to be migrated is not the default base environment (if it is a base environment, use the conda clonecommand to copy it to another name)
Then activate the environment to be migrated. My environment here is named chatGLM, so the corresponding activation command isconda activate chatGLM
Then use the pip command to installpip install conda-pack
Pack the original environment
Use conda pack -n <环境名>to package the environment (the packaging speed is quite fast), and a tar.gz file will be generated after packaging.
It only took me 3 minutes to pack up the environment that I had installed for a day.
Upload the virtual machine environment to the local machine
Note: Both environments must be the same system. For example, my local computer is centos, and the target host is also centos (if your local computer is window, the installed package may not be applicable to centos, which will cause it to not work properly after migration)
Although my native machine is Widnows, I used Virtual box to build a centos virtual machine, installed anaconda in the virtual machine, and then used pip to install the respective dependencies.
Option 1: Upload sz directly
File upload/download commands supported by Tabby terminal
- sz: ZModem protocol to transfer files from Linux server to local
- rz: To upload files from local to Linux server, you can use the rz command
Use sz <待拉取的文件路径>to pull files to local
Option 2: After splitting and slicing, upload sz
Note: The file is relatively large. The sz command may report an error after all transmissions are completed. You can use splitthe command to split it.
split -b 512M -d -a 1 chatGLM.tar.gz
- -b 4096M The size of each compressed package divided is 4G
- The -d parameter specifies that the suffix of the generated segmented package is in the form of a number.
- -ax to set the length of the sequence (the default value is 2), here the length of the sequence is set to 1
After switching to 512M, it can be uploaded, but occasionally an error will still be reported, which is very inconvenient. Although wildcards are supported, you will still be asked for the folder you want to save every time (you cannot set it once, and the folder will be used by default later, which is inconvenient)
Solution three: Virtualbox installs auxiliary plug-ins and shared folders
Virtual cannot directly install the enhanced assistant, even after referring to several blogs, I succeeded after following this article. The address is: https://zhuanlan.zhihu.com/p/633877183
Except for Chinese garbled characters, English is normal (using Tabby or XShell to connect, there will be no garbled characters problem)
Then use mvthe command to copy the file to the shared directory. After copying, switch to the window side and see that it appears instantly (much faster than the previous sz command)
Upload the local environment compressed package to the target server
Tabby then switches to the target server and rzuploads the file to the server using the command
Delete the environment with the same name on the target server
Now come to the target host and use conda env listto view all current environments
Then use conda env remove -p <环境路径>to delete the environment with the same name
restore environment
Then conda env listlist the paths to all virtual environments using
Here, the environment variable path of anaconda3 is:/opt/anaconda3/envs/
Create a new folder with "virtual environment name" under this pathmkdir /opt/anaconda3/envs/<虚拟环境名>
All permissions will then be givenchmod 777 /opt/anaconda3/envs/<虚拟环境名>
Then decompress the compressed package just uploaded to this directory and use the following command
tar -xzvf <环境变量压缩包>.tar.gz -C /opt/anaconda3/envs/<虚拟环境名>
View Results
After modification, use conda env listto view the environment just added
Then use conda activate <虚拟环境名>to activate the environment variable
Use conda listthe command to check whether all the libraries we installed before exist.
