shaffhausen/OpenThinker-32B
OpenThinker-32B is a 32.8 billion parameter language model fine-tuned by shaffhausen based on Qwen/Qwen2.5-32B-Instruct. It is specifically optimized for advanced reasoning and mathematical tasks, demonstrating strong performance on benchmarks like MATH500 and GPQA Diamond. The model leverages the OpenThoughts-114k dataset, distilled from DeepSeek-R1, to enhance its analytical capabilities.
Loading preview...
OpenThinker-32B: Enhanced Reasoning Model
OpenThinker-32B is a 32.8 billion parameter model developed by shaffhausen, fine-tuned from Qwen/Qwen2.5-32B-Instruct. Its primary distinction lies in its training on the unique OpenThoughts-114k dataset, which is derived by distilling DeepSeek-R1. This specialized training focuses on improving the model's performance in complex reasoning and mathematical problem-solving.
Key Capabilities
- Advanced Reasoning: Achieves 90.6 on MATH500 and 61.6 on GPQA Diamond, indicating strong capabilities in mathematical and general reasoning tasks.
- Open-Source Ecosystem: The project emphasizes full transparency, with publicly available model weights, datasets, data generation code, evaluation code (Evalchemy), and training code (LLaMA-Factory).
- Benchmarked Performance: Evaluated using the open-source Evalchemy tool, showing competitive results against other 32B models in reasoning-focused benchmarks.
Good For
- Research and Development: Ideal for researchers and developers focusing on improving reasoning abilities in large language models, given its transparent and open-source nature.
- Mathematical and Scientific Applications: Suitable for use cases requiring strong analytical and problem-solving skills, particularly in mathematics and complex question answering.
- Comparative Analysis: Provides a robust baseline for comparing reasoning performance against other models, especially those trained on proprietary datasets.