HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-010

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-010 model is a 4 billion parameter Qwen3-based instruction-tuned causal language model, developed by HYU-NLP-EVAL. This research checkpoint is specifically optimized through 10 static-rubric RL optimizer updates, focusing on HealthBench prompts. It is designed for studying proxy-rubric staleness during policy optimization, offering insights into reward model behavior rather than direct medical application.

Loading preview...

Model Overview

This model, HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-010, is a research checkpoint derived from the Qwen3-4B-Instruct-2507 base model. It represents the 10th step in a static-rubric Reinforcement Learning (RL) optimization process, specifically part of the static-rubric R0 experiment.

Key Characteristics

  • Base Model: Qwen/Qwen3-4B-Instruct-2507.
  • Optimization: Underwent 10 static-rubric RL optimizer updates, where training rewards utilized a frozen, prompt-specific rubric bank (R0(x)).
  • Training Focus: Policy training was conducted on 256 HealthBench prompts, with full-model RL parameterization.
  • Research Purpose: Primarily intended for studying proxy-rubric staleness during policy optimization, offering insights into the dynamics of reward models.

Intended Use and Limitations

  • Research Tool: This model is a research checkpoint for academic study, particularly in the field of RL optimization and reward model behavior.
  • Not for Medical Use: It is explicitly stated that this model is not a medical device and should not be used as a substitute for professional medical advice. Its static-rubric reward improvement does not guarantee improvement against independent HealthBench ground truth.

Loading Example

from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "HYU-NLP-EVAL/qwen3-4b-healthbench-static-r0-step-010"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id, torch_dtype="bfloat16")

Finetunes
1 models