On June 6, at the Wenxin Large-scale Model Technology Exchange Conference (Chengdu), Baidu Smart Cloud launched the "Comate" code assistant based on large models , and officially opened invitations for testing.
According to reports, with the help of the understanding and reasoning capabilities of Wenxin's large model, "Comate" can quickly complete codes, recommend codes in natural language, automatically find code errors, and comprehensively improve the efficiency of developers' research and development. In the future, developers can use the "Comate" code assistant in mainstream development software through plug-ins and other forms.
▲ Comate multi-line code recommendation
Baidu Smart Cloud said that auxiliary code writing is the main product function in the first phase, and the current "Comate" code assistant can partially realize the writing of code through natural language, completely changing the way of human-computer interaction and program development mode. Baidu insiders said that in the second stage, "Comate" will realize natural language code generation in specific fields and scenarios, and in the third stage, it will realize natural language development in all fields.
Hou Zhenyu, vice president of Baidu Group, said that the "Comate" code assistant is a new generation of coding assistance tools based on large models, and a large number of tests have been conducted within Baidu. The test results show that nearly 50% of the suggested codes written by “Comate” have been adopted by developers, and they have been widely used in various product development within Baidu.
It is said that "Comate" has covered more than 30 programming languages, especially in C/C++, Python, Java, Go, PHP, JavaScript and other mainstream languages. In addition, "Comate" also supports mainstream IDEs most commonly used by programmers. Developers can use "Comate" in different software through plug-ins and other forms. At the same time, combined with the Fei Paddle deep learning framework and Wenxin large model, "Comate" can ensure that a single request for reasoning takes about 300ms, and guarantees user experience with extremely fast response speed.
In the future, Comate will also go online to perform code understanding and description, comment generation, unit test generation, defect identification and automatic repair, code refactoring optimization, automatic stub debugging and other capabilities through selected code blocks, as well as interactive chat through Chat. Capabilities such as natural language generation of codes and instructions, technical knowledge questions and answers, and the ability to directly generate applications or products in natural language in some scenarios, lowering the threshold for software development and improving the quality and efficiency of R&D codes.