Mosaic-IT：面向指令微调的零成本组合式数据合成
Mosaic-IT: Cost-Free Compositional Data Synthesis for Instruction Tuning
AI总结：
本文提出 Mosaic-IT，一种无需人工或教师模型的组合式数据合成方法，通过随机拼接多条指令并引入高层元指令训练大语言模型，在多个基准上提升指令遵循能力，同时将训练成本降低 80%。
摘要：
使用多样化的指令-响应对微调大语言模型，增强了模型理解和遵循指令的能力。当前的指令微调主要依赖教师模型或人工干预来生成和优化训练所需的指令与响应，这种方式成本高昂、不可持续，并且可能缺乏多样性。本文提出 Mosaic Instruction Tuning（Mosaic-IT），一种无需人工或教师模型的组合式数据合成方法，能够从现有指令微调数据中高效生成丰富多样的增强数据，以提升大语言模型的能力。Mosaic-IT 将多条指令数据随机拼接为一条，并训练模型依据预定义的高层元指令生成相应响应，从而强化其多步指令遵循和格式遵循能力。大量评估表明，Mosaic-IT 具有优越的性能和训练效率，在多个基准上取得一致的性能提升，并且与原始指令微调相比，训练成本降低 80%。代码和数据已开源：https://github.com/tianyi-lab/Mosaic-IT。
英文摘要：
Finetuning large language models with a variety of instruction-response pairs has enhanced their capability to understand and follow instructions. Current instruction tuning primarily relies on teacher models or human intervention to generate and refine the instructions and responses for training, which are costly, non-sustainable, and may lack diversity. In this paper, we introduce Mosaic Instruction Tuning (Mosaic-IT), a human/model-free compositional data synthesis method that can efficiently create rich and diverse augmentations from existing instruction tuning data to enhance the LLMs. Mosaic-IT randomly concatenates multiple instruction data into one and trains the model to produce the corresponding responses with predefined higher-level meta-instructions to strengthen its multi-step instruction-following and format-following skills. Our extensive evaluations demonstrate a superior performance and training efficiency of Mosaic-IT, which achieves consistent performance improvements over various benchmarks and an 80% reduction in training costs compared with original instruction tuning. Our codes and data are available at https://github.com/tianyi-lab/Mosaic-IT.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
