MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models

2023/09/21 by Longhui Yu, Weisen Jiang, Yu, Longhui +17 · 100 citations
Computer Science · #Topic Modeling #Natural Language Processing Techniques #Text Readability and Simplification

paper · pdf · doi:10.48550/arxiv.2309.12284

Abstract

Large language models (LLMs) have pushed the limits of natural language understanding and exhibited excellent problem-solving ability. Despite the great success, most existing open-source LLMs (e.g., LLaMA-2) are still far away from satisfactory for solving mathematical problem due to the complex reasoning procedures. To bridge this gap, we propose MetaMath, a fine-tuned language model that specializes in mathematical reasoning. Specifically, we start by bootstrapping mathematical questions by rewriting the question from multiple perspectives without extra knowledge, which results in a new dataset called MetaMathQA. Then we fine-tune the LLaMA-2 models on MetaMathQA. Experimental results on two popular benchmarks (i.e., GSM8K and MATH) for mathematical reasoning demonstrate that MetaMath outperforms a suite of open-source LLMs by a significant margin. Our MetaMath-7B model achieves 66.4% on GSM8K and 19.4% on MATH, exceeding the state-of-the-art models of the same size by 11.5% and 8.7%. Particularly, MetaMath-70B achieves an accuracy of 82.3% on GSM8K, slightly better than GPT-3.5-Turbo. We release all the MetaMathQA dataset, the MetaMath models with different model sizes and the training code for public use.

Cited by

• The Intruder Threshold: A Spectral Law for LoRA Fine-Tuning
• Unifying Learning Dynamics and Generalization in Transformers Scaling Law
• Gamayun's Path to Multilingual Mastery: Cost-Efficient Training of a 1.5B-Parameter LLM
• Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
• Reasoning Palette: Modulating Reasoning via Latent Contextualization for Controllable Exploration for (V)LMs
• Hard Negative Sample-Augmented DPO Post-Training for Small Language Models
• OpenDataArena: A Fair and Open Arena for Benchmarking Post-Training Dataset Value
• Greedy Alignment Principle for Optimizer Selection
• When Do Symbolic Solvers Enhance Reasoning in Large Language Models?
• Making Dialogue Grounding Data Rich: A Three-Tier Data Synthesis Framework for Generalized Referring Expression Comprehension
• Parameter-Efficient Subspace Optimization for LLM Fine-Tuning
• Elastic Mixture of Rank-Wise Experts for Knowledge Reuse in Federated Fine-Tuning
• ORION: Teaching Language Models to Reason Efficiently in the Language of Thought
• LocateAnything3D: Vision-Language 3D Detection with Chain-of-Sight
• HERMES: Towards Efficient and Verifiable Mathematical Reasoning in LLMs
• Tailored Primitive Initialization is the Secret Key to Reinforcement Learning
• Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis for Large Reasoning Models
• EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models
• DynaAct: Large Language Model Reasoning with Dynamic Action Spaces
• Knowledge-Augmented Long-CoT Generation for Complex Biomolecular Reasoning
• DyKAF: Dynamical Kronecker Approximation of the Fisher Information Matrix for Gradient Preconditioning
• NVIDIA Nemotron Nano V2 VL
• Curriculum Design for Trajectory-Constrained Agent: Compressing Chain-of-Thought Tokens in LLMs
• LoRAQuant: Mixed-Precision Quantization of LoRA to Ultra-Low Bits
• Are Language Models Efficient Reasoners? A Perspective from Logic Programming
• Autodata: An agentic data scientist to create high quality synthetic data
• SPICE: Self-Play In Corpus Environments Improves Reasoning
• LoRA-DA: Data-Aware Initialization for Low-Rank Adaptation via Asymptotic Analysis
• SelecTKD: Selective Token-Weighted Knowledge Distillation for LLMs
• ScaLoRA: Optimally Scaled Low-Rank Adaptation for Efficient High-Rank Fine-Tuning
• Foundation of Intelligence: Review of Math Word Problems from Human Cognition Perspective
• Boosting Accuracy and Efficiency of Budget Forcing in LLMs via Reinforcement Learning for Mathematical Reasoning
• DictPFL: Efficient and Private Federated Learning on Encrypted Gradients
• AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data Generation
• SoK: Taxonomy and Evaluation of Prompt Security in Large Language Models
• What Makes a Good Curriculum? Disentangling the Effects of Data Ordering on LLM Mathematical Reasoning
• FineVision: Open Data Is All You Need
• StreamingThinker: Large Language Models Can Think While Reading
• QueST: Incentivizing LLMs to Generate Difficult Problems
• Interpretability Framework for LLMs in Undergraduate Calculus
• Dual-Weighted Reinforcement Learning for Generative Preference Modeling
• A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning
• CarBoN: Calibrated Best-of-N Sampling Improves Test-time Reasoning
• From Pixels to Words -- Towards Native Vision-Language Primitives at Scale
• GroundedPRM: Tree-Guided and Fidelity-Aware Process Reward Modeling for Step-Level Reasoning
• Can MLLMs Absorb Math Reasoning Abilities from LLMs as Free Lunch?
• InteractiveOmni: A Unified Omni-modal Model for Audio-Visual Multi-turn Dialogue
• A Survey on Evaluation of Large Language Models
• Enhancing LLM Reasoning via Non-Human-Like Reasoning Path Preference Optimization
• Enhancing Large Language Model Reasoning via Selective Critical Token Fine-Tuning
• AMiD: Knowledge Distillation for LLMs with α-mixture Assistant Distribution
• Skill-Targeted Adaptive Training
• StelLA: Subspace Learning in Low-rank Adaptation using Stiefel Manifold
• Prompting Test-Time Scaling Is A Strong LLM Reasoning Data Augmentation
• RefGrader: Automated Grading of Mathematical Competition Proofs using Agentic Workflows
• AILoRA: Function-Aware Asymmetric Initialization for Low-Rank Adaptation of Large Language Models
• Mid-Training of Large Language Models: A Survey
• KaVa: Latent Reasoning via Compressed KV-Cache Distillation
• POME: Post Optimization Model Edit via Muon-style Projection
• Making Mathematical Reasoning Adaptive
• LaDiR: Latent Diffusion Enhances LLMs for Text Reasoning
• Scaling Code-Assisted Chain-of-Thoughts and Instructions for Model Reasoning
• EvolProver: Advancing Automated Theorem Proving by Evolving Formalized Problems via Symmetry and Difficulty
• Graph2Eval: Automatic Multimodal Task Generation for Agents via Knowledge Graphs
• Learning a Zeroth-Order Optimizer for Fine-Tuning LLMs
• Train Large, Deploy Compact: Structured Compression for Compact Low-Rank Adaptation
• Communication-Efficient and Accurate Approach for Aggregation in Federated Low-Rank Adaptation
• Thinking-Free Policy Initialization Makes Distilled Reasoning Models More Effective and Efficient Reasoners
• Atomic Thinking of LLMs: Decoupling and Exploring Mathematical Reasoning Abilities
• MixtureVitae: Open Web-Scale Pretraining Dataset With High Quality Instruction and Reasoning Data Built from Permissive-First Text Sources
• Plan before Solving: Problem-Aware Strategy Routing for Mathematical Reasoning with LLMs
• ReasonCACHE: Teaching LLMs To Reason Without Weight Updates
• Beyond English-Centric Training: How Reinforcement Learning Improves Cross-Lingual Reasoning in LLMs
• Don't Settle Too Early: Self-Reflective Remasking for Diffusion Language Models
• Emergence of Superposition: Unveiling the Training Dynamics of Chain of Continuous Thought
• From Deferral to Learning: Online In-Context Knowledge Distillation for LLM Cascades
• Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts
• Blockwise Hadamard high-Rank Adaptation for Parameter-Efficient LLM Fine-Tuning
• SFT Doesn't Always Hurt General Capabilities: Revisiting Domain-Specific Fine-Tuning in LLMs
• Mixture of Thoughts: Learning to Aggregate What Experts Think, Not Just What They Say
• ScaleDiff: Scaling Difficult Problems for Advanced Mathematical Reasoning
• GPO: Learning from Critical Steps to Improve LLM Reasoning
• TDRM: Smooth Reward Models with Temporal Difference for LLM RL and Inference
• Accelerate Scaling of LLM Finetuning via Quantifying the Coverage and Depth of Instruction Set
• Delta Activations: A Representation for Finetuned Large Language Models
• Implicit Reasoning in Large Language Models: A Comprehensive Survey
• Blockwise SFT for Diffusion Language Models: Reconciling Bidirectional Attention and Autoregressive Decoding
• Fin-PRM: A Domain-Specialized Process Reward Model for Financial Reasoning in Large Language Models
• Beyond Pass@1: Self-Play with Variational Problem Synthesis Sustains RLVR
• Mini-Omni-Reasoner: Token-Level Thinking-in-Speaking in Large Speech Models
• Can Large Models Teach Student Models to Solve Mathematical Problems Like Human Beings? A Reasoning Distillation Method via Multi-LoRA Interaction
• Utilizing Multilingual Encoders to Improve Large Language Models for Low-Resource Languages
• InternBootcamp Technical Report: Boosting LLM Reasoning with Verifiable Task Scaling
• Sample-efficient LLM Optimization with Reset Replay
• Unveiling Over-Memorization in Finetuning LLMs for Reasoning Tasks
• CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge
• SAND-Math: Using LLMs to Generate Novel, Difficult and Useful Mathematics Questions and Answers
• CoLD: Counterfactually-Guided Length Debiasing for Process Reward Models in Mathematical Reasoning
• MathDuels: Evaluating LLMs as Problem Posers and Solvers
• Mono-InternVL-1.5: Towards Cheaper and Faster Monolithic Multimodal Large Language Models

Related

• SkyMath: Technical Report
• MathGenie: Generating Synthetic Data with Question Back-translation for Enhancing Mathematical Reasoning of LLMs
• Common 7B Language Models Already Possess Strong Math Capabilities
• MathFimer: Enhancing Mathematical Reasoning by Expanding Reasoning Steps through Fill-in-the-Middle Task
• Can LLMs Master Math? Investigating Large Language Models on Math Stack Exchange