Intelligent-Internet/II-Medical-7B-Preview

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Apr 28, 2025Architecture:Transformer0.0K Cold

II-Medical-7B-Preview by Intelligent-Internet is a 7.6 billion parameter medical reasoning model, fine-tuned from Qwen/Qwen2.5-7B-Instruct. It is specifically optimized for medical question answering and reasoning tasks, leveraging a comprehensive dataset of medical knowledge. The model demonstrates strong performance across various medical QA benchmarks, making it suitable for enhancing AI capabilities in the medical domain.

Loading preview...

II-Medical-7B-Preview: Medical Reasoning Model

II-Medical-7B-Preview is a 7.6 billion parameter language model developed by Intelligent-Internet, specifically designed for medical reasoning. It is built upon the Qwen/Qwen2.5-7B-Instruct architecture and has undergone extensive fine-tuning using a proprietary medical knowledge dataset. The training involved both Supervised Fine-Tuning (SFT) and DAPO (Direct Advantage Policy Optimization) on hard-reasoning medical data to boost performance.

Key Capabilities

  • Specialized Medical Reasoning: Excels in medical question answering across diverse benchmarks like MedMCQA, MedQA, PubMedQA, and medical-related MMLU-Pro and GPQA.
  • Comprehensive Training Data: Trained on over 555,000 samples, including public medical reasoning datasets, synthetic medical QA data, and curated medical R1 traces.
  • Robust Decontamination: Employs a two-step decontamination pipeline, including 10-gram and fuzzy decontamination, to ensure evaluation integrity.

Good For

  • Developing AI applications requiring advanced medical reasoning capabilities.
  • Research and development in medical AI, particularly for question answering systems.

Limitations

  • The dataset may contain inherent biases from source materials.
  • Medical knowledge requires regular updates; the model's knowledge base is static post-training.
  • Not suitable for direct medical use or clinical decision-making.