HYU-NLP-EVAL/qwen3-4b-rar-medicine-onlinerubrics-seed11-step-027

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

HYU-NLP-EVAL/qwen3-4b-rar-medicine-onlinerubrics-seed11-step-027 is a 4 billion parameter language model, an intermediate policy checkpoint from dynamic OnlineRubrics-Every GRPO training, distinct from static-rubric GRPO. Based on Qwen/Qwen3-4B-Instruct-2507 with thinking disabled, this model represents a historical policy state used by the Phase-1 audit. It is intended for research use only, with no medical capability or safety claims, and is not validated for clinical decision-making.

Loading preview...

Model Overview

This model, HYU-NLP-EVAL/qwen3-4b-rar-medicine-onlinerubrics-seed11-step-027, is a 4 billion parameter language model derived from Qwen/Qwen3-4B-Instruct-2507. It represents an intermediate policy checkpoint from a dynamic OnlineRubrics-Every GRPO training process, which differs from static-rubric GRPO methods. Notably, the 'thinking' capability is disabled in this specific iteration.

Key Characteristics

  • Training Origin: An intermediate policy from dynamic OnlineRubrics-Every GRPO training.
  • Base Model: Built upon Qwen/Qwen3-4B-Instruct-2507.
  • Historical Checkpoint: This specific checkpoint served as a historical policy state during a Phase-1 audit.
  • Research Use Only: The model is explicitly for research purposes; no medical capability or safety claims are made, and it is not validated for clinical decision-making.

Technical Details

  • Parameter Count: 4 billion parameters.
  • Context Length: Supports a context length of 32768 tokens.
  • Precision: Root files are provided as veRL-exported Hugging Face inference model in BF16 format.

Important Note

The original_checkpoint/ directory contains the exact original FSDP parameter checkpoint along with tokenizer and configuration files. Optimizer state, training data, responses, rubrics, infrastructure configuration, and credentials are not included in this release.

Finetunes
1 models