Qwen2.5-Math-1.5B: Specialized Mathematical LLM

Qwen2.5-Math-1.5B is part of the Qwen2.5-Math series, an upgrade to the original Qwen2-Math family, developed by Qwen. This 1.5 billion parameter model is specifically engineered for mathematical problem-solving and supports a substantial context length of 131072 tokens.

Key Capabilities

• Multilingual Math Support: Solves mathematical problems in both English and Chinese.
• Advanced Reasoning: Utilizes both Chain-of-Thought (CoT) and Tool-integrated Reasoning (TIR) to improve reasoning capabilities and computational accuracy.
• Enhanced Performance: Achieves significant performance improvements over the Qwen2-Math series on Chinese and English mathematics benchmarks using CoT.
• Tool Integration: TIR further boosts proficiency in precise computation, symbolic manipulation, and algorithmic reasoning, addressing challenges where CoT alone might fall short.

Intended Use

• Mathematical Problem Solving: Primarily designed for tackling math problems, including complex equations and algorithmic tasks.
• Instruction-tuned Version: The Qwen2.5-Math-1.5B-Instruct variant is available for conversational use and direct instruction following.
• Base Model for Fine-tuning: Qwen2.5-Math-1.5B serves as a robust base model for completion, few-shot inference, and as a starting point for further fine-tuning for specific mathematical applications.

Note: This model is explicitly warned against use for tasks other than solving English and Chinese math problems.