luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-HessianMaskToken-0.001-v3_1520
The luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-HessianMaskToken-0.001-v3_1520 is a 7.6 billion parameter language model, fine-tuned from Qwen/Qwen2.5-Math-7B. It was trained using the GRPO method on the MATH-lighteval dataset, specializing in mathematical reasoning. This model is designed to enhance performance on complex mathematical tasks, leveraging advanced reinforcement learning techniques.
Loading preview...
Overview
This model, luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-HessianMaskToken-0.001-v3_1520, is a 7.6 billion parameter language model fine-tuned from the base Qwen/Qwen2.5-Math-7B model. It leverages the GRPO (Generalized Reinforcement Learning with Policy Optimization) method, as introduced in the DeepSeekMath paper, to significantly improve its mathematical reasoning capabilities. The training was conducted using the TRL framework on the DigitalLearningGmbH/MATH-lighteval dataset, specifically targeting enhanced performance in mathematical problem-solving.
Key Capabilities
- Advanced Mathematical Reasoning: Specialized fine-tuning with GRPO on a dedicated mathematical dataset enhances its ability to understand and solve complex math problems.
- Reinforcement Learning Optimization: Utilizes the GRPO method, a technique proven to push the limits of mathematical reasoning in open language models.
- Qwen2.5 Architecture: Built upon the robust Qwen2.5-Math-7B foundation, providing a strong base for its specialized mathematical skills.
Good for
- Applications requiring high accuracy in mathematical problem-solving.
- Research and development in advanced mathematical reasoning with large language models.
- Tasks that benefit from models optimized with reinforcement learning techniques for specific domains.