schwyzquant/OpenThinker-7B
OpenThinker-7B is a 7.6 billion parameter instruction-tuned language model developed by schwyzquant, fine-tuned from Qwen2.5-7B-Instruct. It is specifically enhanced for reasoning tasks, leveraging the OpenThoughts-114k dataset which distills DeepSeek-R1. This model demonstrates improved performance on various reasoning benchmarks, making it suitable for complex analytical applications.
Loading preview...
OpenThinker-7B: Enhanced Reasoning Model
OpenThinker-7B is a 7.6 billion parameter language model, fine-tuned by schwyzquant from the Qwen2.5-7B-Instruct architecture. Its core differentiator is its training on the extensive OpenThoughts-114k dataset, which is derived by distilling DeepSeek-R1. This specialized training significantly enhances its reasoning capabilities, building upon and outperforming its predecessor, Bespoke-Stratos-7B.
Key Capabilities & Performance
- Advanced Reasoning: Demonstrates improved scores across various reasoning benchmarks, including AIME24, MATH500, GPQA-Diamond, and LCBv2.
- Open-Source Ecosystem: The model, its training data, data generation code, and evaluation tools are all publicly available, fostering transparency and community contributions.
- Efficient Training: Trained using four 8xH100 nodes for 20 hours, utilizing specific hyperparameters detailed in the repository.
Ideal Use Cases
- Complex Problem Solving: Suited for applications requiring strong logical deduction and analytical thinking.
- Research & Development: Provides a robust base for further fine-tuning or experimentation in reasoning-focused AI tasks.
- Benchmarking: Can be used as a competitive baseline for evaluating new reasoning models or datasets.
This model is released under the Apache 2.0 License, with comprehensive resources available on the Open Thoughts GitHub Repository and an accompanying research paper.