HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-030
This model is a 4 billion parameter Qwen3-based causal language model, specifically a research checkpoint from the HYU-NLP-EVAL project. It is fine-tuned using a static-rubric Reinforcement Learning (RL) optimization process, focusing on HealthBench prompts. This checkpoint, at step 30, is designed for studying proxy-rubric staleness during policy optimization rather than for direct medical application.
Loading preview...
Overview
This model, HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-030, is a 4 billion parameter Qwen3-Instruct-based language model. It represents a research checkpoint, specifically pi_30, from an experiment focused on static-rubric Reinforcement Learning (RL) optimization. The training process involved 30 updates using a frozen, prompt-specific rubric bank, where each training prompt was scored with its own R0(x) throughout optimization.
Key Characteristics
- Base Model: Qwen/Qwen3-4B-Instruct-2507.
- Optimization Method: Full-model Reinforcement Learning with a static-rubric (R0) reward source.
- Training Data: Policy training was conducted on 256 HealthBench prompts.
- Purpose: Primarily intended for research into proxy-rubric staleness during policy optimization, rather than as a general-purpose medical AI.
Intended Use and Limitations
This model is a research checkpoint and not a medical device. It should not be used as a substitute for professional medical advice. The improvement observed through static-rubric reward does not inherently guarantee improvement against independent HealthBench ground truth. Developers can load the model using the transformers library with AutoModelForCausalLM and AutoTokenizer.