CodeShell is a code model developed by the Knowledge Computing Laboratory of the National Engineering Research Center for Software Engineering at Peking University in collaboration with the AI Laboratory of Sichuan Tianfu Bank. It is a large-scale model with 7 billion parameters, known as the "most powerful code base of its kind".
CodeShell is based on 500 billion tokens for cold start training, with a context window length of 8192. Its architecture design incorporates the core features of both StarCoder and Llama.
The goal of CodeShell is to provide full-stack intelligent programming support, including code generation, code completion, code correction, code refactoring, code commenting, code documentation, code evaluation, and other functions.
CodeShell supports multiple programming languages, including Python, Java, C#, C++, JavaScript, HTML, CSS, SQL, etc. CodeShell can also generate code based on natural language descriptions or generate natural language descriptions based on code. CodeShell can also perform cross-language conversion, transforming code from one language to another.
How to use CodeShell?
Using CodeShell is very simple, just install the CodeShell plugin in VS Code, and you can directly call CodeShell's functions in the editor.
Many people may directly search for CodeShell in VS Code! But the search results are wrong!
This is wrong! Wrong! Wrong!
The correct way is to follow the steps introduced on the official website and download and install it!
After executing the command, you will see this file in the folder: codeshell-vscode-0.0.1.vsix
Drag this file to the vscode plugin area and it will be automatically installed. After installation, an instruction page will open. This is the correct way!
Then you need to configure the plugin's code autocompletion feature.
By completing these steps, you have installed the plugin, but only halfway. You still need to deploy the large model.
According to the official instructions, after completing these steps, you can run it. However, during the cloning process, the author kept experiencing timeouts and was unsure if it was due to the popularity of the large model and the high number of downloads.
Interested friends can try it themselves and see if they can successfully deploy it!
CodeShell open source repository:
https://github.com/WisdomShell/codeshell
CodeShell's VSCode plugin open source repository:
https://github.com/WisdomShell/codeshell-vscode
CodeShell base model:
https://huggingface.co/WisdomShell/CodeShell-7B
Evaluation open source repository:
https://github.com/WeOpenML/PandaLM