luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-v2_2389
The luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-v2_2389 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 DigitalLearningGmbH/MATH-lighteval dataset, specializing it for mathematical reasoning tasks. This model is optimized to enhance performance in complex mathematical problem-solving, building upon the base Qwen 2.5 architecture.
Loading preview...
Model Overview
This model, luckeciano/Qwen-2.5-7B-DrGRPO-Adam-FisherMaskToken-1e-4-v2_2389, is a fine-tuned version of the Qwen/Qwen2.5-Math-7B base model, featuring 7.6 billion parameters. It leverages the Qwen 2.5 architecture and has been specifically adapted for mathematical reasoning.
Key Differentiator: GRPO Training
The primary distinction of this model lies in its training methodology. It was fine-tuned using GRPO (Generalized Reinforcement Learning with Policy Optimization), a method introduced in the paper "DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models". This technique aims to improve the model's ability to handle complex mathematical problems.
Training Data and Framework
- Dataset: The model was trained on the
DigitalLearningGmbH/MATH-lightevaldataset, which is designed for evaluating mathematical capabilities. - Framework: Training was conducted using the
TRL(Transformer Reinforcement Learning) library.
Intended Use
This model is particularly well-suited for:
- Mathematical reasoning: Solving and explaining mathematical problems.
- Academic research: Exploring the impact of GRPO on language models for math.
- Applications requiring precise numerical and logical deduction.