MASTER: Enhancing Large Language Model via Multi-Agent Simulated Teaching

2025/06/03 by Yihong Tang, Yue, Liang, Tang, Yihong +6
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Intelligent Tutoring Systems and Adaptive Learning #Multimodal Machine Learning Applications #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2506.02689

openalex publication_date 2025/06/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Instruction fine-tuning is crucial in NLP tasks, enhancing pretrained models' instruction-following capabilities and task-specific performance. However, obtaining high-quality fine-tuning data for large models is challenging due to data collection difficulties and high production costs. To address this, we propose MASTER, a novel data augmentation method that enriches original data through interactions among multiple agents with varying cognitive levels. We simulate three pedagogically grounded teaching scenarios, leveraging multi-agent conversations to generate high-quality teacher-student interaction data. Utilizing MASTER, we construct BOOST-QA, a fine-tuning dataset augmented from existing datasets like Orca-Math-200k, ProcQA, and OpenHermes2.5. Experiments show that models fine-tuned with BOOST-QA perform excellently across multiple benchmarks, demonstrating strong multitask generalization. Notably, MASTER significantly improves models' reasoning abilities in complex tasks, providing valuable insights for future research.

Citations

• ORPP: Self-Optimizing Role-playing Prompts to Enhance Language Model Capabilities
• Thinking in Character: Advancing Role-Playing Agents with Role-Aware Reasoning
• TinyR1-32B-Preview: Boosting Accuracy with Branch-Merge Distillation
• Generator-Assistant Stepwise Rollback Framework for Large Language Model Agent
• The Rise of Darkness: Safety-Utility Trade-Offs in Role-Playing Dialogue Agents
• Condor: Enhance LLM Alignment with Knowledge-Driven Data Synthesis and Refinement
• Qwen2.5 Technical Report
• Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling
• The Llama 3 Herd of Models
• TAGCOS: Task-agnostic Gradient Clustered Coreset Selection for Instruction Tuning Data
• Simulating Classroom Education with LLM-Empowered Agents
• MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark
• PGA-SciRE: Harnessing LLM on Data Augmentation for Enhancing Scientific Relation Extraction
• ProCQA: A Large-scale Community-based Programming Question Answering Dataset for Code Search
• Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey
• Unveiling the Truth and Facilitating Change: Towards Agent-based Large-scale Social Movement Simulation
• Orca-Math: Unlocking the potential of SLMs in Grade School Math
• CompeteAI: Understanding the Competition Dynamics in Large Language Model-based Agents
• Mistral 7B
• Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View
• The CoT Collection: Improving Zero-shot and Few-shot Learning of Language Models via Chain-of-Thought Fine-Tuning
• Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
• WizardLM: Empowering large pre-trained language models to follow complex instructions
• Generative Agents: Interactive Simulacra of Human Behavior
• Self-Instruct: Aligning Language Models with Self-Generated Instructions
• Training language models to follow instructions with human feedback
• Program Synthesis with Large Language Models
• FlipDA: Effective and Robust Data Augmentation for Few-Shot Learning
• Evaluating Large Language Models Trained on Code
• Text Data Augmentation for Deep Learning
• GPT3Mix: Leveraging Large-scale Language Models for Text Augmentation
• Measuring Mathematical Problem Solving With the MATH Dataset
• CoDA: Contrast-enhanced and Diversity-promoting Data Augmentation for Natural Language Understanding
• Measuring Massive Multitask Language Understanding
• Similarity-Preserving Knowledge Distillation
• Relational Knowledge Distillation
• EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• Crowdsourcing Multiple Choice Science Questions
• Deep Mutual Learning
• Distilling the Knowledge in a Neural Network

Related

• AgentInstruct: Toward Generative Teaching with Agentic Flows
• EducationQ: Evaluating LLMs' Teaching Capabilities Through Multi-Agent Dialogue Framework
• TeacherLM: Teaching to Fish Rather Than Giving the Fish, Language Modeling Likewise
• Embracing Imperfection: Simulating Students with Diverse Cognitive Levels Using LLM-based Agents
• Language Models as Science Tutors