近日，阿里巴巴旗下通义千问团队推出了一款创新的开源代码大模型——CodeQwen1.5。这款基于Qwen语言模型的专家级代码生成工具，以其强大的功能和广泛的语言支持，再次刷新了业界对开源代码模型的认知。CodeQwen1.5具备70亿参数，能够处理长达64K的上下文输入，涵盖了从基础到复杂的92种编程语言，这在开源领域无疑是一次重大突破。
在性能表现上，CodeQwen1.5在代码生成、长序列建模、代码修改以及SQL能力等多个关键任务中展现出卓越的性能。在HumanEval、MBPP和LiveCodeBench等基准测试中，CodeQwen1.5的优异表现超越了众多同类开源模型，甚至在某些方面与知名的大模型GPT-3.5-Turbo相比也毫不逊色。
这一创新成果的发布，不仅为开发者提供了更强大、更灵活的代码智能辅助工具，也为开源社区注入了新的活力。CodeQwen1.5的推出，标志着阿里巴巴在人工智能和代码智能领域的持续领先地位，同时也预示着未来编程工作将更加高效、智能。我们期待这款开源代码大模型能在实践中不断优化，推动全球软件开发技术的进一步发展。
英语如下：
**News Title:** “Alibaba Breaks New Ground in Open Source: CodeQwen1.5 Supports 92 Programming Languages, Elevating Code Processing to New Heights”
**Keywords:** Alibaba Open Source, CodeQwen1.5, Programming Language Support
**News Content:**
Title: Alibaba’s Qwen Team Launches CodeQwen1.5, Setting a New Milestone in Open Source Code Expert Models
Recently, Alibaba’s Qwen team unveiled CodeQwen1.5, an innovative open-source code large language model. Based on the Qwen language model, this expert-level code generation tool has refreshed industry perceptions with its robust capabilities and extensive language support. CodeQwen1.5, equipped with 7 billion parameters, can handle context inputs as long as 64K, accommodating 92 programming languages ranging from basic to complex – a significant breakthrough in the open-source domain.
In terms of performance, CodeQwen1.5 excels in key tasks such as code generation, long sequence modeling, code modification, and SQL capabilities. Outperforming numerous同类open-source models in benchmarks like HumanEval, MBPP, and LiveCodeBench, CodeQwen1.5’s exceptional performance even rivals that of the well-known GPT-3.5-Turbo in certain aspects.
This groundbreaking release not only provides developers with a more powerful and flexible code intelligence assistant but also infuses new vitality into the open-source community. The launch of CodeQwen1.5 underscores Alibaba’s continued leadership in artificial intelligence and code intelligence, foreshadowing a future of more efficient and intelligent programming. We look forward to the ongoing optimization of this open-source code model as it drives further advancements in global software development technology.
【来源】https://qwenlm.github.io/zh/blog/codeqwen1.5/
Views: 1