**北大开源新代码大模型aiXcoder-7B为企业私有部署而来，领先行业性能显著**
近日，北京大学的aiXcoder团队宣布最新开源代码大模型aiXcoder-7B Base版的推出。该模型专为在企业软件开发场景中部署而设计，旨在解决企业私有环境中的代码智能化需求。
据悉，aiXcoder-7B Base经过在超过1.2T Unique Tokens上的大量训练，其预训练任务和上下文信息针对真实代码生成场景进行了独特设计。在HumanEval、MBPP和MultiPL-E三大主流评测集上的表现尤为突出，平均得分超越了其他同等级参数量模型，如Codellama等。特别是在代码补全场景下的表现卓越，显示出强大的潜力。
此外，在多语言NL2Code基准测试中，aiXcoder 7B Base的效果平均超过了Codellama 34B和StarCoder2 1
进行测评的数据显示它无论是在性能指标还是在适应性方面都有着超越对手的表现。业内人士指出，这一开源大模型的推出将有助于企业实现软件开发的智能化升级，提高开发效率和代码质量。该模型不仅可以缩短开发周期，减少人工错误，还能助力企业构建更加智能的软件应用。
目前，aiXcoder团队已经在GitHub等平台上公开了这一大模型的详细信息及使用方法。企业可通过其官方渠道下载和使用这一开源模型进行软件开发。业界期待其未来的更多突破与创新，以推动智能代码开发的发展进程。
英语如下：
News Title: PKU’s aiXcoder-7B Model: The First Choice for Enterprise Private Deployment, Surpassing Codellama with Amazing Performance!
Keywords: 1. PKU Open Source Model
News Content: **PKU’s Open Source aiXcoder-7B Model Designed for Enterprise Private Deployment, with Significant Industry-leading Performance**
Recently, the aiXcoder team from Peking University announced the launch of the latest open source code model, aiXcoder-7B Base. This model is specifically designed for deployment in enterprise software development scenarios, aiming to address the intelligent code needs in private environments.
It is reported that aiXcoder-7B Base has been extensively trained on more than 1.2T Unique Tokens, and its pre-training tasks and contextual information are uniquely designed for real code generation scenarios. It has particularly impressive performance on the three major evaluation sets of HumanEval, MBPP, and MultiPL-E, with an average score surpassing other models with the same level of parameters, such as Codellama. Its excellent performance in code completion scenarios shows great potential.
In addition, in the multi-language NL2Code benchmark test, the effectiveness of aiXcoder 7B Base averaged over Codellama 34B and StarCoder2 1. Evaluation data shows that it surpasses its competitors in both performance indicators and adaptability. Industry insiders point out that the launch of this open source model will help enterprises achieve intelligent upgrades in software development, improve development efficiency and code quality. This model can not only shorten the development cycle and reduce manual errors, but also help enterprises build more intelligent software applications.
Currently, the aiXcoder team has publicly released the detailed information and usage of this large model on platforms such as GitHub. Enterprises can download and use this open source model for software development through its official channels. The industry is looking forward to more breakthroughs and innovations in the future to promote the development of intelligent code development.
【来源】https://www.qbitai.com/2024/04/134070.html
Views: 1