The fourth generation of CodeGeeX model, as the latest masterpiece of Zhipu AI, has been officially released and open sourced, marking a new milestone in code generation technology. This model demonstrates unparalleled performance and versatility in areas below 10B parameters. Since the first generation of models was launched in 2014, CodeGeeX has been updated every six months, continuously promoting the boundaries of code generation capabilities and providing developers with a more efficient and smarter programming experience.
CodeGeeX not only has basic functions such as code completion, annotation, repair and translation, but also introduces advanced functions such as networked Q&A, code interpreter, and retrieval enhancement to fully cover the needs of developers. For individual users, these features are completely free and support multiple mainstream IDEs, allowing developers to seamlessly integrate into existing workflows. At present, the number of individual users has exceeded 1 million, and the enterprise version is also widely used in many industries and has been highly recognized in the evaluation of the China Academy of Information and Communications Technology.
CodeGeeX4-ALL-9B, the open source version of the CodeGeeX4 series model, integrates a variety of programming and development functions, covering multiple scenarios from code generation to debugging. The model performs well on multiple code review sets, demonstrating its powerful code generation capabilities. In addition, the model supports 128K context, is able to process long code files and achieves 100% accuracy in code retrieval tasks. It is worth mentioning that CodeGeeX4-ALL-9B is the only code model that implements Function Call, with a success rate of more than 90% in function call testing, providing developers with higher reliability.
The CodeGeeX plug-in v2.12.0 version has been fully connected to the fourth generation model, bringing faster code generation speed and multiple functional innovations. For example, automatic generation of project README files, long text context memory, cross-file analysis and Q&A, local mode support, and improvements in NL2SQL capabilities. These features not only enhance the efficiency of code search, understanding and maintenance, but also significantly improve the overall experience of developers.
Developers can easily experience this latest technology by upgrading the CodeGeeX plug-in in the IDE or searching and downloading in the plug-in market. At the same time, Zhipu AI has also released enterprise-level CodeGeeX4 generation software and hardware integrated machine products, providing code warehouse-level rewriting, Q&A and retrieval enhancement, supporting information and innovation solutions, suitable for more enterprise-level application scenarios, further expanding the application of CodeGeeX scope.
Project address: https://github.com/THUDM/CodeGeeX4