**北大最新开源aiXcoder-7B Base代码大模型 专为企业私有部署设计**
近日，北京大学aiXcoder团队最新推出开源代码大模型——aiXcoder-7B Base版，特别为企业软件开发场景打造。该模型拥有7B参数规模，显示出强大的实力。
经过大量训练，该模型在超过1.2T Unique Tokens的数据上表现卓越。它的预训练任务和上下文信息针对真实代码生成环境进行了专门设计，以确保其在企业私有部署中的高效性能。值得一提的是，aiXcoder 7B Base在HumanEval、MBPP和MultiPL-E三大主流评测集上的平均得分超过了Codellama模型。在代码补全场景下的表现尤为突出，其效果在同等级参数量模型中名列前茅。
此外，在多语言NL2Code基准测试中，aiXcoder 7B Base的平均效果也超越了Codellama 34B和StarCoder2 15B等其他主流模型。这表明北京大学的新模型不仅为企业级应用提供了强大的支持，同时也标志着中国在自然语言处理和人工智能领域的进一步发展。预计这一开源模型的推出将极大推动行业的技术创新与应用实践。业内人士分析认为，该模型的高效表现有望改变软件开发行业的现状，引领未来代码生成的趋势。
英语如下：
News Title: PKU’s aiXcoder-7B Model Open-Sourced: A Powerful Tool for Enterprise Private Deployment, Surpassing Competitors with Excellent Performance
Keywords: PKU open-source aiXcoder-7B, enterprise private deployment, surpassing Codellama
News Content: **PKU’s Latest Open-Source aiXcoder-7B Base Code Model Designed for Enterprise Private Deployment**
Recently, the aiXcoder team from Peking University has released the latest open-source code model, aiXcoder-7B Base, specifically designed for enterprise software development scenarios. With a 7B parameter scale, this model demonstrates great strength.
After extensive training, the model exhibits excellent performance on data exceeding 1.2T Unique Tokens. Its pre-training tasks and contextual information are specifically designed for real code generation environments, ensuring efficient performance in enterprise private deployment. Notably, aiXcoder 7B Base outperforms the Codellama model on average in three mainstream evaluation sets, including HumanEval, MBP, and MultiPL-E. It particularly excels in code completion scenarios, ranking among the best in its parameter size class.
Additionally, in the multi-language NL2Code benchmark test, aiXcoder 7B Base surpasses other mainstream models such as Codellama 34B and StarCoder2 15B. This indicates that the new model from Peking University not only provides powerful support for enterprise applications but also marks the further development of China in the fields of natural language processing and artificial intelligence. It is expected that the release of this open-source model will greatly promote technological innovation and application practices in the industry. Industry insiders believe that the efficient performance of this model is expected to change the current situation of the software development industry and lead the trend of future code generation.
【来源】https://ai-bot.cn/go/?url=aHR0cHM6Ly93d3cucWJpdGFpLmNvbS8yMDI0LzA0LzEzNDA3MC5odG1s
Views: 0