bofenghuang/docto-decision-gemma-4-e4b-fr-v0.1
Docto Decision Gemma-4-E4B (v0.1) by bofenghuang is a 7.9 billion parameter French medical decision model, fine-tuned from Google's gemma-4-E4B-it with a LoRA adapter. It specializes in medical decision tasks, providing probability scores for given answers based on patient messages, clinical notes, or case reports. This model excels at tasks like patient triage, specialty routing, and clinical text analysis, achieving 90.13% mean accuracy on the Docto Decision Bench, significantly outperforming zero-shot LLMs in French medical contexts.
Loading preview...
Docto Decision Gemma-4-E4B (v0.1) Overview
Docto Decision Gemma-4-E4B is a specialized French medical decision model developed by bofenghuang. It is a LoRA fine-tune of Google's gemma-4-E4B-it (8.00B parameters), designed to process French medical texts and provide probabilistic answers to specific questions. The model takes a patient message, clinical note, or case report, along with a question and possible answers, then returns a probability for each answer. It is optimized for single-token responses, such as an option letter or "oui" / "non", with probabilities derived from these tokens.
Key Capabilities
- Medical Decision Making: Provides probabilistic answers for various medical scenarios based on textual input.
- French Language Focus: Specifically trained and optimized for French medical contexts.
- Flexible Task Definition: Tasks are defined at inference time via instructions, questions, and options, allowing for zero-shot application to new, similar tasks.
- High Accuracy: Achieves a mean accuracy of 90.13% on the Docto Decision Bench, outperforming other open LLMs and some frontier models in medical decision tasks.
Good for
- Patient Message Triage: Determining urgency levels, identifying red flags, and routing to appropriate specialties.
- Practice Inbox Management: Classifying request intent, assigning tasks, and managing reminders.
- Assistant Safety: Implementing input guardrails, checking answer compliance, and verifying rule adherence.
- Clinical Text Analysis: Extracting assertions, temporality, medications, allergies, and supporting claims.
- Medical Coding and Document Classification: Assigning ICD-10 chapters, identifying document types, and assessing passage relevance.