全球科技巨头 Meta 近日宣布了一项重大进展，其内部研发的代码生成人工智能模型 Code Llama 70B 已经完成更新，成为目前市场上最大、性能最强的模型之一。这一开源工具自去年8月发布以来，已受到广大开发者和研究者的热烈欢迎和广泛应用。
Meta 的首席执行官马克·扎克伯格在公开声明中表示，新版 Code Llama 70B 在处理复杂查询的能力上实现了显著提升，旨在为全球开发者提供更强大、更高效的编程辅助。在关键的 HumanEval 基准测试中，Code Llama 70B 的准确率达到了53%，超越了竞争对手 GPT-3.5 的48.1%。
尽管这一成绩令人瞩目，但 Meta 的新模型与 OpenAI 的论文中提及的 GPT-4 相比，仍有提升空间。据 OpenAI 的数据，GPT-4 在相同基准测试中的准确率高达67%。Meta 表示，将持续优化和改进 Code Llama，以期在未来缩小与 GPT-4 的差距，甚至实现超越。
这一开源项目凸显了 Meta 对于人工智能在代码生成领域的重视，同时也为全球开发者提供了一个免费且强大的工具，有望进一步推动编程自动化和效率的提升。随着 Code Llama 70B 的发布，Meta 在人工智能领域的影响力和贡献再次得到彰显。
英语如下：
**News Title:** “Meta Outpaces GPT-3.5 with Code Llama 70B: The World’s Largest and Most Performant Code-Generating AI Unveiled”
**Keywords:** Meta Code Llama, 70B model, performance boost
**News Content:**
### Meta Launches Open-Source Code Llama 70B, Outperforming GPT-3.5
Global tech giant Meta recently announced a major advancement with the update of its in-house code-generating AI model, Code Llama 70B, now becoming one of the largest and most powerful models in the market. Since its release last August, this open-source tool has been warmly welcomed and widely adopted by developers and researchers worldwide.
Mark Zuckerberg, Meta’s CEO, stated in a public announcement that the new Code Llama 70B has shown significant improvements in handling complex queries, aiming to provide developers globally with more robust and efficient programming assistance. In the crucial HumanEval benchmark test, Code Llama 70B achieved an accuracy of 53%, surpassing GPT-3.5’s 48.1%.
Despite these impressive results, Meta’s new model still lags behind OpenAI’s mentioned GPT-4. According to OpenAI data, GPT-4 scores 67% on the same benchmark. Meta has expressed its commitment to continuously optimizing and enhancing Code Llama to narrow the gap with GPT-4 and potentially surpass it in the future.
This open-source project underscores Meta’s emphasis on AI in code generation and offers developers worldwide a free and powerful tool, poised to further advance automation and efficiency in programming. With the release of Code Llama 70B, Meta’s influence and contribution in the AI domain are once again highlighted.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1