SeongryongJung/qwen3-4b-material-rlsd-ema005

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The SeongryongJung/qwen3-4b-material-rlsd-ema005 model is a 4 billion parameter language model fine-tuned from Qwen/Qwen3-4B. It utilizes RLSD (Reinforcement Learning from Scientific Data) with an EMA of 0.05, specifically optimized on the 'material' split of the SciKnowEval dataset. This model demonstrates strong performance in scientific knowledge evaluation, achieving a peak mean@16 score of 77.19% on validation metrics, making it suitable for tasks requiring scientific reasoning and material science understanding.

Loading preview...

Model Overview

SeongryongJung/qwen3-4b-material-rlsd-ema005 is a 4 billion parameter language model derived from the Qwen3-4B architecture. It has undergone a specialized fine-tuning process using Reinforcement Learning from Scientific Data (RLSD) with an Exponential Moving Average (EMA) of 0.05. The training specifically targeted the 'material' split of the SciKnowEval dataset, indicating an optimization for scientific and material-related knowledge.

Key Capabilities

  • Scientific Knowledge Evaluation: The model is fine-tuned to excel in tasks related to scientific understanding, particularly within the domain of materials science.
  • RLSD Optimization: The use of RLSD aims to enhance the model's ability to reason and generate accurate responses based on scientific data.
  • Performance on SciKnowEval: Achieved a peak validation performance of 77.19% mean@16 on the SciKnowEval reward metric, demonstrating its proficiency in this specialized area.

Good For

  • Material Science Applications: Ideal for use cases requiring deep understanding or generation of content related to material science.
  • Scientific Reasoning Tasks: Suitable for tasks that benefit from a model trained on scientific datasets and optimized for factual accuracy in scientific contexts.
  • Research and Development: Can serve as a foundational model for further research or application development in scientific domains, especially where Qwen3-4B's base capabilities are enhanced by scientific fine-tuning.

Finetunes
1 models