SeongryongJung/Qwen3-4B-Material-GRPO-TR

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 3, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

SeongryongJung/Qwen3-4B-Material-GRPO-TR is a 4 billion parameter Qwen3-based language model fine-tuned using the GRPO method. This model is specifically optimized for material science knowledge, achieving a 76.60% mean@16 score on the Material / SciKnowEval material dataset. It features a 32768 token context length and is designed for specialized applications requiring deep understanding in material science.

Loading preview...

Overview

SeongryongJung/Qwen3-4B-Material-GRPO-TR is a specialized 4 billion parameter language model built upon the Qwen3 architecture. It has been fine-tuned using the GRPO (Generalized Reinforcement Learning with Policy Optimization) method, specifically targeting the domain of material science. The training process involved a batch size of 32 and focused on the Material / SciKnowEval material dataset.

Key Capabilities

  • Material Science Expertise: Achieves a peak validation mean@16 score of 76.60% on the Material / SciKnowEval material dataset, indicating strong performance in this specialized domain.
  • GRPO Fine-tuning: Utilizes the GRPO method for optimization, with detailed hyperparameters including a learning rate of 1e-6 and a total of 100 training steps.
  • Context Length: Supports a maximum prompt length of 2048 tokens and a maximum response length of 8192 tokens, with a total model length of 10240 tokens.

Good For

  • Specialized Material Science Tasks: Ideal for applications requiring deep knowledge and reasoning within the material science field.
  • Research and Development: Suitable for researchers and developers working on material-related NLP problems who need a model with domain-specific fine-tuning.
  • Benchmarking GRPO: Can serve as a reference for evaluating the effectiveness of the GRPO fine-tuning method on domain-specific datasets.