Technology Radar
Published : Apr 03, 2024
NOT ON THE CURRENT EDITION
This blip is not on the current edition of the Radar. If it was on one of the last few editions, it is likely that it is still relevant. If the blip is older, it might no longer be relevant and our assessment might be different today. Unfortunately, we simply don't have the bandwidth to continuously review blips from previous editions of the Radar.
Understand more
Apr 2024
Hold
许多组织都在试图将大语言模型(LLMs)应用于他们的产品、领域或组织知识,我们看到了太多 急于冲向大语言模型微调(fine-tune LLMs) 的情况。虽然这种操作的确可以强大到对特定任务的用例进行优化,但在许多情况下对大语言模型进行微调并不是必需的。最常见误用是为了让 LLM 应用程序了解特定的知识、事实或组织的代码库进行微调。在绝大多数场景下,使用检索增强生成(RAG)可以提供更好的解决方案和更优的投入产出比。微调需要大量的计算资源和专家能力,并且比 RAG 面临更多敏感和专有数据挑战。此外当你没有足够的数据进行微调时,还有欠拟合(underfitting)的风险。又或者,当你拥有太多数据时(这倒不太常见),出现过拟合(overfitting)风险。总之达到你所需要任务专业性的正确平衡是比较困难的。在你急于为应用场景进行大语言模型微调前,需要仔细考虑这些权衡和替代方案。