bofenghuang/docto-decision-qwen3.5-4b-fr-v0.1
The bofenghuang/docto-decision-qwen3.5-4b-fr-v0.1 model is a 4.5 billion parameter LoRA fine-tune of Qwen3.5-4B, specifically designed for French medical decision tasks. It processes patient messages, clinical notes, or case reports, a question, and possible answers, returning a probability for each answer. This model excels in use cases like patient triage, practice inbox management, assistant safety checks, and clinical text analysis, demonstrating high accuracy on the Docto Decision Bench.
Loading preview...
Docto Decision Qwen3.5-4B (v0.1) Overview
This model is a specialized 4.5 billion parameter language model, fine-tuned from Qwen3.5-4B using LoRA, to perform French medical decision tasks. It takes a text input (e.g., patient message, clinical note), a question, and a set of possible answers, then outputs a probability for each answer. The model is designed to provide single-token responses, representing an option letter or "oui" / "non", with probabilities derived from these tokens.
Key Capabilities
- French Medical Decision-Making: Optimized for various medical decision tasks in French, including patient triage, urgency level assessment, and specialty routing.
- Probabilistic Answers: Provides a probability score for each potential answer, allowing for nuanced decision support.
- Flexible Task Definition: Tasks are defined at inference time through instructions, questions, and answer options, supporting zero-shot application to new, similar tasks.
- High Accuracy: Achieves 88.94% mean accuracy on the Docto Decision Bench, outperforming several other models in its class for medical decision tasks.
Good For
- Patient Message Analysis: Determining urgency levels, identifying red flags, and routing messages to appropriate specialties.
- Practice Inbox Management: Categorizing request intent, identifying who should respond, and managing visit motives.
- Assistant Safety: Implementing input guardrails, ensuring answer compliance, and performing rule/rubric checks.
- Clinical Text Processing: Asserting temporality, identifying medications and allergies, and supporting claim documentation.
- Medical Coding and Document Classification: Assisting with ICD-10 chapter classification, document type identification, and passage relevance.
Limitations
This is a research model and not a medical device; human oversight is crucial. It may underestimate urgency and is weaker on tasks far from its training data or general medical knowledge questions. Its training labels are primarily derived from the Qwen model family, inheriting some of its conventions and potential errors. The model is exclusively French-language.