Meta 公司近期宣布推出新版代码生成 AI 模型 Code Llama 70B，该模型在性能上超越了 GPT-3.5。Meta 首席执行官扎克伯格表示，Code Llama 70B 是迄今为止公司开发的“最大、性能最优的模型”。这项工具最初于去年8月发布，旨在为所有人提供免费使用的代码生成服务。
新版 Code Llama 70B 在处理查询的能力上有了显著提升。在 HumanEval 基准测试中，该模型的准确率达到了 53%，超过了 GPT-3.5 的 48.1%。然而，与 OpenAI 论文中提到的 GPT-4 相比，Code Llama 70B 在准确性上还有提升空间，GPT-4 的准确率达到了 67%。
Title: Meta Unveils Code Llama 70B: Outperforms GPT-3.5 in Performance
Keywords: Meta, Code Llama, GPT-3.5
News content:
Meta has recently announced the launch of its latest AI code-generating model, Code Llama 70B. The model outperforms GPT-3.5 in terms of performance. Mark Zuckerberg, CEO of Meta, stated that Code Llama 70B is the “largest and most capable model” developed by the company. The tool was first released last August and is available for free to everyone.
The updated Code Llama 70B has significantly improved its ability to handle queries. It achieved an accuracy rate of 53% on the HumanEval benchmark, surpassing GPT-3.5’s 48.1%. However, it still has room for improvement compared to GPT-4, as mentioned in an OpenAI paper, which has an accuracy rate of 67%.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1