OctoMed/OctoMed-7B
OctoMed-7B is a 7 billion parameter multimodal medical reasoning model developed by Timothy Ossowski, Sheng Zhang, and others. Built upon Qwen2.5-VL-7B-Instruct, it is fine-tuned on a large-scale dataset of over 8 million structured reasoning traces and 6.8 billion response tokens, distilled from DeepSeek-R1 and GPT-4o. This model excels at robust clinical reasoning across various out-of-distribution medical benchmarks, producing internal reasoning traces before its final answer. It supports a 32768 token context length and is optimized for medical visual question answering and diagnostic support.
Loading preview...
OctoMed-7B: High-Performance Multimodal Medical Reasoning
OctoMed-7B is a 7 billion parameter multimodal model specifically designed for advanced medical reasoning. It is built on the Qwen2.5-VL-7B-Instruct base model and has been extensively fine-tuned using a unique, large-scale dataset.
Key Capabilities & Features
- Specialized Medical Reasoning: Developed through a scalable data pipeline that distills structured reasoning traces from powerful models like DeepSeek-R1 and GPT-4o.
- Extensive Training Data: Trained on the largest multimodal medical reasoning dataset to date, comprising over 8 million traces and 6.8 billion response tokens.
- Robust Benchmark Performance: Achieves strong and consistent performance across a wide array of out-of-distribution medical benchmarks, including VQA-RAD and SLAKE.
- Internal Reasoning Traces: Generates internal thought processes (within
<think>...</think>tokens) before providing a final answer, with longer traces for more complex queries. - Multimodal Input: Supports image and text inputs, making it suitable for medical visual question answering (VQA).
- Optimized for Multiple-Choice VQA: Specifically fine-tuned for multiple-choice VQA tasks, with recommended prompting strategies for optimal results.
When to Use OctoMed-7B
- Medical Image Analysis: Ideal for tasks involving medical image interpretation and related question answering.
- Clinical Decision Support: Can assist in generating reasoned responses for clinical queries based on multimodal input.
- Research in Medical AI: A valuable tool for researchers exploring advanced medical AI applications and reasoning.
Note: The model is sensitive to system prompts; the default system prompt ("You are a helpful assistant.") is recommended. It is primarily fine-tuned for multiple-choice VQA, and while it may follow instructions for other tasks, it has not been extensively tested for them.