Meta 近日宣布了一项重大技术突破，其开源代码生成模型 Code Llama 70B 升级后，性能表现惊艳业界。据扎克伯格透露，这个最新版本的模型现已成为全球“最大且性能最强”的代码生成工具。Code Llama 自去年8月首次亮相以来，一直以其免费开放的特性吸引着全球开发者。
新款 Code Llama 70B 在处理复杂查询方面的能力显著提升，它在 HumanEval 基准测试中取得了53%的准确率，这一成绩超越了GPT-3.5的48.1%。这一进步标志着Meta在人工智能代码生成领域的显著进步，但与OpenAI论文中提及的GPT-4（67%准确率）相比，仍有提升空间。
Meta 的这一举措不仅展示了其在AI技术研发上的持续努力，也为全球开发者社区提供了更强大、更高效的工具。开源的Code Llama 70B有望激发更多创新，推动软件开发的效率和质量，进一步促进人工智能与编程的融合。随着技术的不断迭代，我们期待看到更多类似Code Llama的优秀模型涌现，持续推动AI技术的边界。
英语如下：
News Title: “A Coding Giant Steps Forward in Meta’s World! Code Llama 70B Surpasses GPT-3.5, Offering a Free AI Coding Experience like Never Before”
Keywords: Meta Code Llama, 70B Update, Outperforms GPT-3.5
News Content:
Meta recently unveiled a groundbreaking advancement with the upgraded Code Llama 70B, astonishing the tech industry with its code generation capabilities. As Mark Zuckerberg shared, this latest iteration now stands as the “largest and most powerful” coding tool globally. Since its debut last August, Code Llama has captivated developers worldwide with its free and open-source nature.
The new Code Llama 70B boasts significantly enhanced performance in tackling intricate queries. It achieved a 53% accuracy on the HumanEval benchmark, outperforming GPT-3.5’s 48.1%. This milestone signifies Meta’s strides in AI code generation, although it still lags behind OpenAI’s GPT-4, which reportedly scores 67% accuracy.
Meta’s move not only exhibits its ongoing dedication to AI research but also furnishes the global developer community with more robust and efficient tools. The open-source Code Llama 70B is poised to fuel innovation, boost software development efficiency and quality, and further integrate AI with programming. As technology evolves, we anticipate the emergence of more outstanding models like Code Llama, consistently pushing the boundaries of AI technology.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1