FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview
| Property | Value |
|---|---|
| Parameter Count | 32B |
| Model Type | Merged Language Model |
| Author | FuseAI Team |
| Implementation | Long-Long Reasoning Merge |
What is FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview?
FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview represents a breakthrough in AI model fusion, combining the strengths of DeepSeek-R1-Distill-Qwen-32B, QwQ-32B-Preview, and Sky-T1-32B-Preview. This innovative model achieves remarkable performance in System-II reasoning tasks, particularly excelling in mathematics, coding, and scientific reasoning domains.
Implementation Details
The model employs advanced SCE merging methodologies to integrate multiple open-source LLMs into a unified architecture. It specifically utilizes Long-Long Reasoning Merging, focusing on enhancing long-chain-of-thought reasoning capabilities across different model architectures.
- Achieves 74.0% Pass@1 accuracy on AIME24
- Demonstrates 86.7% Cons@32 performance
- Surpasses OpenAI's o1-preview (44.6%) and o1-mini (63.4%)
- Shows strong performance in MATH500 (94.8%) and OlympiadBench (65.0%)
Core Capabilities
- Advanced mathematical reasoning and problem-solving
- Enhanced scientific reasoning across multiple domains
- Strong performance in coding tasks (57.9% on LiveCodeBench)
- Improved consistency in complex problem-solving
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its fusion of multiple high-performing LLMs, specifically optimized for System-II reasoning tasks. It demonstrates superior performance compared to individual models while approaching the capabilities of OpenAI's o1 in certain benchmarks.
Q: What are the recommended use cases?
The model excels in mathematical problem-solving, scientific reasoning, and coding tasks. It's particularly well-suited for applications requiring detailed step-by-step reasoning and complex problem-solving in academic and technical domains.
