SeongryongJung/Qwen3-4B-Material-RLSD-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-RLSD-TR is a 4 billion parameter Qwen3-based language model fine-tuned by SeongryongJung using the RLSD_TR method. This model is specifically optimized for material science knowledge, achieving 76.86% on the SciKnowEval material dataset. It is designed for tasks requiring specialized understanding and generation within the domain of materials science, leveraging a 32768 token context length.

Loading preview...

Model Overview

SeongryongJung/Qwen3-4B-Material-RLSD-TR is a specialized 4 billion parameter language model built upon the Qwen3-4B architecture. It has been fine-tuned using the RLSD_TR (Reinforcement Learning with Self-Distillation and Trust-Region) method, specifically targeting the domain of material science.

Key Capabilities and Performance

  • Material Science Specialization: The model demonstrates strong performance on the SciKnowEval material dataset, achieving a validation mean@16 score of 76.86% after 100 training steps.
  • RLSD_TR Fine-tuning: Utilizes a sophisticated reinforcement learning approach with a trust-region policy for improved stability and performance during fine-tuning.
  • Context Length: Supports a maximum model length of 10240 tokens, with a max prompt length of 2048 and max response length of 8192, enabling processing of substantial material science texts.
  • Training Details: Trained with a batch size of 32 over 100 steps, focusing on the datasets/sciknoweval/material dataset.

Ideal Use Cases

This model is particularly well-suited for applications requiring deep understanding and generation of content related to material science. Potential use cases include:

  • Scientific Text Analysis: Extracting information or summarizing research papers in material science.
  • Knowledge Retrieval: Answering specific questions about material properties, compositions, or processes.
  • Educational Tools: Assisting students and researchers with material science concepts.
  • Specialized Content Generation: Creating technical descriptions or reports within the material science domain.

Finetunes
1 models