【北京大学开源创新，aiXcoder-7B代码大模型引领企业级应用新高度】
近日，北京大学的aiXcoder团队在人工智能领域取得了重大突破，他们开源了一款专为企业私有部署设计的代码大模型——aiXcoder-7B Base版。这款模型以其卓越的性能和针对企业软件开发场景的定制化设计，引起了业界的广泛关注。
aiXcoder-7B Base模型在1.2T Unique Tokens的数据集上进行了大规模训练，确保了其在处理复杂代码任务时的高效性和准确性。模型的预训练任务和上下文信息均针对真实的代码生成场景进行了精心设计，以满足企业级开发的实际需求。
在性能评测上，aiXcoder-7B Base在HumanEval、MBPP和MultiPL-E三大主流评测集上的平均得分超越了拥有340亿参数的Codellama模型。特别是在代码补全这一关键场景中，aiXcoder-7B Base展现出了同类参数量模型中的最佳效果。不仅如此，该模型在多语言NL2Code基准测试中，平均表现也优于Codellama 34B和StarCoder2 15B，充分展示了其在跨语言代码理解和生成方面的强大能力。
这一创新成果的开源，不仅彰显了北京大学在人工智能领域的科研实力，也为企业级软件开发提供了更智能、更高效的新工具，有望推动整个行业的发展。北京大学aiXcoder团队的这一举措，无疑将为全球企业带来更先进的代码生成和优化解决方案，加速企业数字化转型的步伐。
英语如下：
**News Title:** “Peking University’s Open-Source Masterpiece: aiXcoder-7B, a Code Generation Model Breaking Records in Enterprise Deployment”
**Keywords:** aiXcoder-7B, enterprise deployment, code generation
**News Content:**
**[Peking University Innovates with aiXcoder-7B, a Large Code Model Redefining Enterprise-Level Applications]**
Recently, the aiXcoder team at Peking University made a significant breakthrough in the field of artificial intelligence by open-sourcing aiXcoder-7B Base, a code generation model specifically designed for enterprise private deployment. The model’s exceptional performance and tailored design for corporate software development scenarios have attracted widespread attention.
Trained on a massive dataset of 1.2T Unique Tokens, the aiXcoder-7B Base model ensures high efficiency and accuracy when handling complex coding tasks. Its pre-training tasks and contextual information have been meticulously crafted to cater to real-world code generation scenarios in enterprise development.
In performance evaluations, the aiXcoder-7B Base outperformed Codellama, a model with 340 billion parameters, on the three major benchmark sets — HumanEval, MBPP, and MultiPL-E. Notably, it achieved the best results among models of similar parameter sizes in the crucial task of code completion. Furthermore, it demonstrated superior average performance on the multi-language NL2Code benchmark compared to Codellama 34B and StarCoder2 15B, highlighting its robust cross-language code understanding and generation capabilities.
By open-sourcing this innovative achievement, Peking University showcases its research prowess in artificial intelligence and provides enterprises with a smarter, more efficient tool for software development, potentially advancing the entire industry. This initiative by the aiXcoder team is set to offer advanced code generation and optimization solutions to global businesses, accelerating their digital transformation journeys.
【来源】https://www.qbitai.com/2024/04/134070.html
Views: 2