sail/Qwen2.5-Math-1.5B-Oat-Zero

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kPublished:Mar 17, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

sail/Qwen2.5-Math-1.5B-Oat-Zero is a 1.5 billion parameter language model developed by sail, fine-tuned for mathematical reasoning tasks. Based on the Qwen2.5-Math-1.5B architecture, it utilizes the minimalist R1-Zero training recipe with the Dr. DRPO algorithm on level 3-5 questions from the MATH dataset. This model is specifically optimized for solving complex mathematical problems, demonstrating its capabilities on widely used math benchmarks.

Loading preview...

Qwen2.5-Math-1.5B-Oat-Zero: Specialized Mathematical Reasoning Model

This model, developed by sail, is a 1.5 billion parameter language model specifically trained for advanced mathematical problem-solving. It is built upon the Qwen2.5-Math-1.5B base model and employs the R1-Zero training recipe with the Dr. DRPO algorithm.

Key Capabilities

  • Mathematical Reasoning: Excels at solving complex math problems, particularly those found in the MATH dataset (levels 3-5).
  • Specialized Training: Utilizes a minimalist training approach, focusing on efficient learning for mathematical tasks.
  • Qwen-Math Template: Designed to work effectively with the Qwen-Math prompting template for optimal performance.

Performance

The model's performance on various math benchmarks is detailed in the associated paper and benchmark table, showcasing its proficiency in mathematical domains.

Good For

  • Applications requiring precise mathematical problem-solving.
  • Research into efficient training methodologies for specialized LLMs.
  • Educational tools for advanced mathematics.