【Meta开源新成果：Code Llama 70B代码生成AI模型超越GPT-3.5】全球科技巨头Meta近日宣布了一项重大进展，其代码生成人工智能模型Code Llama 70B已进行全面升级，被誉为当前“最大、性能最佳”的模型。这一开源项目自去年8月推出以来，已引起广泛关注，而现在，它正以前所未有的能力吸引业界的注意。
Meta首席执行官马克·扎克伯格透露，新版Code Llama 70B在处理复杂查询的能力上显著提升，其在HumanEval基准测试中的准确率达到了53%，这一成绩超越了GPT-3.5的48.1%。这一提升标志着Meta在AI代码生成领域的重大突破，表明Code Llama 70B在理解和生成代码的精度上迈出了坚实的一步。
然而，尽管Code Llama 70B的表现令人印象深刻，但与OpenAI论文中提及的GPT-4（准确率67%）相比，仍存在一定的差距。Meta表示，将持续致力于AI技术研发，以缩小与领先水平的差距，并为全球开发者提供更加强大和精准的工具。
Code Llama 70B的开源特性使得全球的开发者都能免费试用和研究这一先进模型，有望推动全球软件开发效率的提升和创新。这一举措再次证明了Meta在开放源代码和促进技术普惠方面的承诺，同时也预示着AI在代码生成领域将有更加广泛和深入的应用前景。
英语如下：
**News Title:** “Meta Outshines GPT-3.5: Code Llama 70B, an Open-Source Code Generation Model, Achieves Higher Accuracy and is Now Free to Use!”
**Keywords:** Meta Code Llama, 70B model, accuracy improvement
**News Content:** **Meta Unveils Major Upgrade with Code Llama 70B, Outperforming GPT-3.5 in Code Generation AI** Tech giant Meta recently announced a significant advancement in its code-generating AI model, Code Llama 70B, now hailed as the “largest and most performant” model of its kind. Since its launch last August, the open-source project has drawn substantial attention, and it is now captivating the industry with unprecedented capabilities.
Mark Zuckerberg, Meta’s CEO, revealed that the updated Code Llama 70B demonstrates a notable improvement in handling complex queries, achieving a 53% accuracy on the HumanEval benchmark, surpassing GPT-3.5’s 48.1%. This milestone signifies a major breakthrough for Meta in AI code generation, signifying a solid step forward in Code Llama 70B’s precision for understanding and generating code.
However, despite Code Llama 70B’s impressive performance, it still lags behind GPT-4, as mentioned in OpenAI’s papers, which scores 67% accuracy. Meta commits to continuous AI research and development to bridge this gap and provide more powerful and precise tools for global developers.
The open-source nature of Code Llama 70B enables developers worldwide to experiment with and study this advanced model free of charge, potentially boosting global software development efficiency and fostering innovation. This move underscores Meta’s dedication to open-source initiatives and promoting technological inclusivity while foreshadowing a broader and deeper application landscape for AI in code generation.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1