AceMath-7B-Instruct: Advanced Mathematical Reasoning

AceMath-7B-Instruct is a 7.6 billion parameter model from NVIDIA, specifically engineered for high-performance mathematical reasoning. It is developed from the Qwen2.5-Math-7B-Base model through a multi-stage supervised fine-tuning (SFT) process, first with general-purpose SFT data, followed by math-specific SFT data.

Key Capabilities

• Mathematical Problem Solving: Excels at solving complex English mathematical problems using Chain-of-Thought (CoT) reasoning.
• Benchmark Performance: Outperforms previous best-in-class models like Qwen2.5-Math-7B-Instruct on various math reasoning benchmarks, achieving an average pass@1 of 67.2.
• Specialized Training: Benefits from a dedicated math-specific SFT phase, leveraging comprehensive training data released by NVIDIA.

Good For

• Dedicated Math Applications: Primarily recommended for tasks requiring robust mathematical problem-solving capabilities.
• Research in Mathematical AI: The accompanying training data and reward models (AceMath-7B-RM) support further research in this domain.

It's important to note that while highly specialized for math, for general-purpose tasks encompassing code, math, and general knowledge, NVIDIA also offers the AceInstruct series of models.