Kerassy/Qwen3.5-4B-Medical-Reasoning
Kerassy/Qwen3.5-4B-Medical-Reasoning is a 4-billion parameter model fine-tuned from Qwen/Qwen3.5-4B, specifically optimized for complex clinical reasoning and medical question-answering. It leverages chain-of-thought (CoT) fine-tuning to generate explicit step-by-step diagnostic reasoning within tags. The model achieves 68.34% accuracy on the MedQA USMLE benchmark, outperforming many larger non-reasoning models, making it suitable for AI-assisted clinical reasoning research and evaluation.
Loading preview...
Kerassy/Qwen3.5-4B-Medical-Reasoning: Specialized Clinical Reasoning Model
This model, fine-tuned from Qwen/Qwen3.5-4B, is a 4-billion parameter language model expertly designed for complex clinical reasoning and medical question-answering. Its core differentiator is the integration of chain-of-thought (CoT) reasoning, which allows it to generate explicit, step-by-step diagnostic thought processes within <think> tags before providing a final medical conclusion.
Key Capabilities & Features
- Explainable Reasoning: Utilizes CoT fine-tuning on medical reasoning datasets to produce transparent, step-by-step diagnostic thinking, preventing superficial answers.
- Dual Output Modes: Supports a default "Thinking Mode" with CoT traces and a "Direct Mode" for succinct answers by pre-filling
<think>\n</think>tags. - Strong Medical Performance: Achieves 68.34% overall accuracy on the MedQA (USMLE 4-Options) benchmark, demonstrating competitive performance against larger non-reasoning models.
- Efficient Fine-Tuning: Trained using Unsloth (LoRA fine-tuning) on a single NVIDIA A100 GPU, leveraging the
FreedomIntelligence/medical-o1-reasoning-SFTdataset.
When to Use This Model
- Medical Research: Ideal for evaluating chain-of-thought capabilities in compact LLMs within medical contexts.
- AI-Assisted Clinical Reasoning Analysis: Suitable for benchmarking and analyzing AI's ability to perform clinical reasoning.
Important Note: This model is for research and evaluation only and should not be used for direct patient diagnosis or clinical decision-making.