社交巨头Meta近日宣布，对其开源代码生成AI模型Code Llama 70B进行更新，号称是迄今为止“最大、性能最优的模型”。Code Llama工具于去年8月发布，所有人都可以免费使用。
新版Code Llama 70B现在能够处理比之前更多地查询。在HumanEval基准测试中的准确率达到了53%，表现优于GPT-3.5的48.1%，但距离OpenAI论文中给出的GPT-4准确率（67%）还有一段差距。
这次更新的Code Llama 70B模型的性能提升，展示了Meta在人工智能领域的强大实力。据悉，Code Llama 70B模型参数达到了700亿，这使得它在处理大规模任务时具有明显优势。
尽管Code Llama 70B在准确率上超过了GPT-3.5，但仍需进一步提高才能与GPT-4相媲美。业内人士表示，Meta此次发布的Code Llama 70B模型将对人工智能领域的研究和发展产生重要影响。
英文标题：Meta Unveils Open Source Large Model Code Llama 70B With Higher Accuracy Than GPT-3.5
英文关键词：Meta, Code Llama, Artificial Intelligence
英文新闻内容：
Social media giant Meta recently announced an update to its open-source code generation AI model, Code Llama 70B, which it claims is “the largest and most performant model to date.” The Code Llama tool was released last August and is available for free to everyone.
The updated Code Llama 70B can now handle more queries than before. It achieved an accuracy of 53% on the HumanEval benchmark test, outperforming GPT-3.5’s 48.1%, but there is still a gap to bridge to match the accuracy of GPT-4 reported in OpenAI’s paper, which stands at 67%.
This performance improvement of the Code Llama 70B model demonstrates Meta’s strong capabilities in the field of artificial intelligence. It is said that the Code Llama 70B model has 70 billion parameters, which gives it a significant advantage in handling large-scale tasks.
Despite surpassing GPT-3.5 in accuracy, the Code Llama 70B still needs to be further improved to match the capabilities of GPT-4. Industry insiders believe that the release of the Code Llama 70B model by Meta will have a significant impact on the research and development of the artificial intelligence field.
【来源】https://www.ithome.com/0/747/917.htm
Views: 0