AI-MO/NuminaMath-72B-CoT
AI-MO/NuminaMath-72B-CoT is a 72 billion parameter language model developed by AI-MO, fine-tuned from Qwen/Qwen2-72B. It specializes in solving competition-level math problems using Chain of Thought (CoT) reasoning. This model is the first stage of a two-stage fine-tuning process, trained on over 860,000 math problem-solution pairs from the AI-MO/NuminaMath-CoT dataset. It is designed for mathematical reasoning tasks, particularly those requiring step-by-step problem-solving.
Loading preview...
NuminaMath 72B CoT: Specialized Mathematical Reasoning
NuminaMath 72B CoT is a 72 billion parameter language model developed by AI-MO, specifically engineered for solving competition-level mathematics problems. Fine-tuned from Qwen/Qwen2-72B, this model represents the first stage of a two-stage supervised fine-tuning process, focusing on Chain of Thought (CoT) reasoning.
Key Capabilities & Training:
- Mathematical Problem Solving: Trained extensively on the AI-MO/NuminaMath-CoT dataset, comprising over 860,000 math competition problem-solution pairs, with solutions templated for CoT.
- CoT Reasoning: Designed to generate step-by-step reasoning for complex math problems, facilitating transparent and verifiable solutions.
- Performance: Capable of solving problems at the level of AMC 12, demonstrating strong performance in its specialized domain.
Intended Uses & Limitations:
- Primary Use Case: Ideal for applications requiring advanced mathematical problem-solving, particularly those found in math competitions.
- Domain Specificity: This model is highly specialized for mathematics and is not intended for general chat applications.
- Current Limitations: While proficient at AMC 12 level, it may struggle with harder problems (e.g., AIME, Math Olympiad) and geometry problems due to its current capacity and lack of multimodal capabilities.