paper-with-me

홈 › Papers

Mosaic-IT: Free Compositional Data Augmentation Improves Instruction Tuning

2024-05-22 · Ming Li, Pei Chen, Chenguang Wang, Hongyu Zhao, Yijun Liang, Yupeng Hou, Fuxiao Liu, Tianyi Zhou

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 augmentation 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 a $80\%$ reduction in training costs compared with original instruction tuning. Our codes and data are available at https://github.com/tianyi-lab/Mosaic-IT.

📄 PDF Abstract BibTeX arXiv:2405.13326

Code (3)

tianyi-lab/mosaic-it 공식 구현
mingliiii/ruler
tianyi-lab/ruler

Tasks

Data AugmentationDiversityInstruction Following

Similar Papers 제목 키워드 기반

Memory Mosaics

2024-05-10 · Jianyu Zhang, Niklas Nolte, Ranajoy Sadhukhan, Beidi Chen 외

Memory Mosaics are networks of associative memories working in concert to achieve a prediction task of interest. Like transformers, memory mosaics possess compositional capabilities and in-context learning capabilities. …

DisentanglementIn-Context LearningLanguage ModelingLanguage Modelling

Select-Mosaic: Data Augmentation Method for Dense Small Object Scenes

2024-06-08 · Hao Zhang, Shuaijie Zhang, Renbin Zou

Data augmentation refers to the process of applying a series of transformations or expansions to original data to generate new samples, thereby increasing the diversity and quantity of the data, effectively improving the…

Data AugmentationDiversityobject-detectionObject Detection+1

Mosaic Representation Learning for Self-supervised Visual Pre-training

2023-02-28 · journal 2023 2 · Zhaoqing Wang, Ziyu Chen, Yaqian Li, Yandong Guo 외

Self-supervised learning has achieved significant success in learning visual representations without the need for manual annotation. To obtain generalizable representations, a meticulously designed data augmentation stra…

Data AugmentationRepresentation LearningSelf-Supervised Learning

Mosaic: Compositional Multi-Concept Erasure via Vector Field Blending

2026-05-25 · Junseok Ko, Jungwoo Kim, Jong-Seok Lee arxiv

Concept erasure has emerged as a key research direction for ensuring safe and ethical image synthesis in Text-to-Image (T2I) models. While existing studies have explored concept erasure across multiple concepts, they typ…

AudioMosaic: Contrastive Masked Audio Representation Learning

2026-05-14 · Hanxun Huang, Qizhou Wang, Xingjun Ma, Cihang Xie 외 arxiv

Audio self-supervised learning (SSL) aims to learn general-purpose representations from large-scale unlabeled audio data. While recent advances have been driven mainly by generative reconstruction objectives, contrastive…

Self-Supervised LearningRepresentation LearningContrastive Learning