HYU-NLP-EVAL/qwen3-4b-rar-medicine-static-r0-matched-seed11-step-039
The HYU-NLP-EVAL/qwen3-4b-rar-medicine-static-r0-matched-seed11-step-039 is a 4-billion parameter language model based on the Qwen3 architecture, fine-tuned for the medical domain. It is an intermediate research checkpoint from a matched static-rubric GRPO (Generative Reinforcement Learning with Policy Optimization) run, specifically trained on the RaR-Medicine dataset with 1,500 prompts. This model is optimized for generating responses within a medical context, utilizing a static reward function, and is not intended for clinical use.
Loading preview...
Overview
This model, HYU-NLP-EVAL/qwen3-4b-rar-medicine-static-r0-matched-seed11-step-039, is a 4-billion parameter Qwen3-based language model. It represents an intermediate research checkpoint from a Generative Reinforcement Learning with Policy Optimization (GRPO) experiment, specifically a matched static-rubric run. The model has undergone 39 global optimizer updates and is distinct from dynamic-rubric checkpoints.
Key Capabilities
- Medical Domain Specialization: Fine-tuned on the RaR-Medicine dataset, making it suitable for tasks requiring medical knowledge.
- GRPO Training: Developed using a static-rubric GRPO method, focusing on reward signals derived from
rar_static_r0_only. - Research Checkpoint: Provides a snapshot of a policy during an ongoing research experiment, allowing for analysis of training progression.
Good for
- Medical Text Generation Research: Ideal for researchers exploring the application of LLMs in the medical domain, particularly those interested in reinforcement learning approaches.
- Understanding GRPO: Useful for studying the effects of static-rubric GRPO training on a Qwen3-4B model within a specialized domain.
- Non-Clinical Prototyping: Can be used for developing and testing applications that require medical-themed text generation, provided no clinical claims or safety-critical uses are intended.