Saved in:
Bibliographic Details
Main Authors: Li, Ming, Chen, Pei, Wang, Chenguang, Zhao, Hongyu, Liang, Yijun, Hou, Yupeng, Liu, Fuxiao, Zhou, Tianyi
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.13326
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913875654344704
author Li, Ming
Chen, Pei
Wang, Chenguang
Zhao, Hongyu
Liang, Yijun
Hou, Yupeng
Liu, Fuxiao
Zhou, Tianyi
author_facet Li, Ming
Chen, Pei
Wang, Chenguang
Zhao, Hongyu
Liang, Yijun
Hou, Yupeng
Liu, Fuxiao
Zhou, Tianyi
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2405_13326
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Mosaic-IT: Cost-Free Compositional Data Synthesis for Instruction Tuning
Li, Ming
Chen, Pei
Wang, Chenguang
Zhao, Hongyu
Liang, Yijun
Hou, Yupeng
Liu, Fuxiao
Zhou, Tianyi
Computation and Language
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.
title Mosaic-IT: Cost-Free Compositional Data Synthesis for Instruction Tuning
topic Computation and Language
url https://arxiv.org/abs/2405.13326