DeepMath-1.5B: Advanced Mathematical Reasoning

DeepMath-1.5B is a 1.5 billion parameter model developed by zwhe99, specifically engineered for high-level mathematical reasoning. It is built upon the deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B architecture and has been fine-tuned using Reinforcement Learning (RL) on the unique DeepMath-103K dataset.

Key Capabilities & Features

• Specialized Mathematical Reasoning: Optimized for solving complex mathematical problems, particularly those at Levels 5-9 difficulty.
• Diverse Mathematical Coverage: Addresses a broad spectrum of topics including Algebra, Calculus, Number Theory, Geometry, Probability, and Discrete Mathematics.
• Novel and Decontaminated Data: Trained on DeepMath-103K, a dataset featuring novel problems rigorously decontaminated against common benchmarks to ensure fair evaluation and minimize test set leakage.
• Rich Data Format: Leverages a dataset where each problem includes a question, a verifiable final answer, difficulty score, topic classification, and multiple reasoning paths (R1 Solutions) for robust training.
• State-of-the-Art Performance: Achieves competitive results on challenging math benchmarks, as indicated by its SOTA performance claims.

Good for

• Developing AI for advanced mathematics: Ideal for applications requiring strong problem-solving capabilities in various math domains.
• Research in mathematical reasoning: Provides a robust base for exploring and improving AI's ability to understand and solve complex math problems.
• Educational tools: Can be integrated into systems designed to assist with or generate challenging math exercises.