s1.1-32B
| Property | Value |
|---|---|
| Author | SimpleScaling |
| Model Size | 32B parameters |
| Paper | Research Paper |
| Repository | Hugging Face |
What is s1.1-32B?
s1.1-32B is an advanced language model that succeeds the original s1-32B, specifically designed to enhance reasoning capabilities by leveraging reasoning traces from r1 instead of Gemini. The model represents a significant improvement in mathematical reasoning and problem-solving capabilities, trained on a carefully curated dataset of 1K examples.
Implementation Details
The model implements budget forcing techniques, specifically ignoring end-of-thinking and appending "Wait" up to four times to optimize performance. It builds upon the architecture of its predecessor while incorporating improved reasoning patterns through r1 traces, developed in collaboration with Bespoke Labs.
- Trained on 1,000 high-quality examples
- Implements advanced budget forcing techniques
- Utilizes r1 reasoning traces for enhanced performance
Core Capabilities
- MATH500 Performance: 95.4% accuracy
- AIME2024 Score: 56.7%
- AIME2025 I Score: 60.0%
- GPQA-Diamond: 63.6% accuracy
- Significantly improved reasoning capabilities compared to predecessor
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its improved reasoning capabilities achieved through the integration of r1 traces, showing particular strength in mathematical problem-solving tasks with significant improvements over its predecessor in tests like AIME2025 I.
Q: What are the recommended use cases?
Given its strong performance on mathematical reasoning tasks, the model is particularly well-suited for advanced mathematical problem-solving, educational applications, and scenarios requiring complex logical reasoning.
