基本信息 - How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition

arxiv How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition

阅读

Star 0

名称: How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition

首页: https://yiyibooks.cn/arxiv/2310.05492v3/index.html

原始地址: https://arxiv.org/pdf/2310.05492.pdf

描述

具有大量预训练标记和参数的大型语言模型 (LLM) 会出现多种能力，包括数学推理、代码生成和指令跟踪。这些能力通过监督微调（SFT）得到进一步增强。虽然开源社区已经探索了专门的 SFT 来增强个人能力，但专有的 LLM 却表现出了跨各种技能的多功能性 ...