New competitor AI assistant for programmers, Ant Group announces CodeFuse, a large open source code model

According to news on September 8, there are already many AI products on the market that can assist in writing codes. At today’s Bund Conference sub-forum, Ant Group open sourced the large code model CodeFuse for the first time. Join this market competition.

According to reports, this is a large code generation model independently developed by Ant. It can provide intelligent suggestions and real-time support based on developer input, Help developers automatically Generate code, automatically add comments, automatically generate test cases, repair and optimize code, etc. to improve R&D efficiency.

Ant Group said that whether you are a beginner or an experienced developer, CodeFuse can greatly improve programming efficiency and accuracy, making programming and innovation a reality for everyone

CodeFuse is based on Ant basic large model development, in the recent HumanEval evaluation of code completion, CodeFuse scored 74.4%, exceeding the score of GPT-4 (67%), and also exceeding the score of WizardCoder-34B of 73.2%. This open source content includes code frameworks, models, etc., which are now available on relevant platforms. Partners of this site can download the models at

HuggingFace.

The code framework part includes the self-developed multi-task fine-tuning (MFT) framework.

MFT supports code generation, code translation, test case generation, bug fixing and other 10 tasks to be fine-tuned together. The MFT fine-tuning framework supports Ant's self-developed models and code capability fine-tuning for multiple open source large models. At the same time, the framework contains innovative fine-tuning data packaging technology, which is claimed to increase fine-tuning speed by about 8 times; and uses multiple deployment optimization technology to accelerate inference by about 2 times.

The model part includes two large models, CodeFuse13B-4K and CodeFuse-CodeLlaMa34B-MFT. During the model training process, CodeFuse used program analysis and verification technology to screen high-quality code data and built a dictionary dedicated to the code field, with the average coding length being 10% lower.

Ant Group revealed that CodeFuse began internal testing as early as June this year. It is used in a real development environment and can help developers with low-threshold programming, such as writing H5 mini-games in natural language, quickly developing Alipay mini-programs, etc. . Application scenarios based on CodeFuse include development assistants, IDE plug-ins, data analyzers, etc., covering the main needs of current research and development work, and

have been gradually verified in Ant Group’s internal research and development process.

The above is the detailed content of New competitor AI assistant for programmers, Ant Group announces CodeFuse, a large open source code model. For more information, please follow other related articles on the PHP Chinese website!