ncbi/Med-V1-Q3B
Med-V1-Q3B is a 3.1 billion parameter small language model developed by ncbi, fine-tuned from Qwen2.5-3B-Instruct with a 32768 token context length. It is specifically optimized for zero-shot and scalable biomedical evidence attribution, excelling at verifying assertions against scientific articles. This model provides an efficient and accurate alternative to larger frontier LLMs for tasks requiring biomedical claim verification and hallucination detection.
Loading preview...
Overview of Med-V1-Q3B
Med-V1-Q3B is a 3.1 billion parameter language model developed by ncbi, specifically designed for biomedical evidence attribution. It is fine-tuned from Qwen2.5-3B-Instruct using a high-quality synthetic dataset called MedFact-Synth. This model's primary function is to assess how strongly a given scientific article supports or refutes a specific assertion, providing a score from -2 (Strong Contradiction) to +2 (Strong Agreement).
Key Capabilities
- Biomedical Evidence Attribution: Accurately verifies claims against provided source articles in a zero-shot manner.
- Scalable Solution: Offers a lightweight and efficient alternative to larger, more expensive frontier LLMs like GPT-5 for biomedical verification tasks.
- High Performance: Substantially outperforms its base models (by +27.0% to +71.3%) on five biomedical benchmarks and performs comparably to frontier LLMs despite its smaller size.
- Explanation Generation: Provides high-quality, step-by-step explanations for its attribution predictions.
- Hallucination Detection: Can quantify hallucinations in LLM-generated answers and identify high-stakes misattributions in clinical guidelines.
Important Considerations
- Context-Dependent Classification: Med-V1-Q3B classifies support or refutation based only on the provided source evidence, not on universal factual validity. A factually true claim might be refuted if the article presents conflicting data, and vice-versa.
- Not for Medical Diagnosis: The model's output is not intended for direct diagnostic use or medical decision-making without professional review.