近日，Meta 发布了一款名为 Code Llama 70B 的人工智能大模型，该模型在代码生成领域的准确率表现优异。据该公司首席执行官扎克伯格透露，Code Llama 70B 是迄今为止“最大、性能最优的模型”。相较于前版本，Code Llama 70B 能够处理更多的查询，在 HumanEval 基准测试中的准确率达到了 53%，超过了 GPT-3.5 的 48.1%。然而，这一成绩仍略逊于 OpenAI 论文中给出的 GPT-4 准确率（67%）。
Meta 开源的 Code Llama 70B 模型为所有人免费使用，旨在推动人工智能技术的发展和应用。尽管 Code Llama 70B 在准确率上仍有提升空间，但其表现已足以证明 Meta 在人工智能领域的实力和投入。
英文翻译：
News Title: Meta Releases Open-source Large Model Code Llama 70B, Outperforming GPT-3.5
Keywords: Meta, Code Llama, GPT-3.5, Artificial Intelligence
News Content:
Recently, Meta released an artificial intelligence large model named Code Llama 70B, which demonstrates excellent accuracy in the field of code generation. According to CEO Zuckerberg, Code Llama 70B is the “largest and highest-performance model” to date. Compared to the previous version, Code Llama 70B can handle more queries and achieve a 53% accuracy rate in HumanEval benchmark tests, outperforming GPT-3.5’s 48.1%. However, its performance still lags behind the GPT-4 accuracy rate of 67% reported in OpenAI’s paper.
The open-source Code Llama 70B model is available for free to everyone, aiming to promote the development and application of artificial intelligence technology. Although there is still room for improvement in Code Llama 70B’s accuracy, its performance demonstrates Meta’s strength and commitment in the field of artificial intelligence.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1