Edifon/SOAP_SFT_V1: Medical SOAP Note Generator
SOAP_SFT_V1 is a 4.3 billion parameter language model developed by Edifon, fine-tuned from unsloth/gemma-3-4b-it-unsloth-bnb-4bit. Its primary function is to automatically generate structured clinical SOAP notes (Subjective, Objective, Assessment, Plan) from doctor-patient dialogues, making it a specialized tool for clinical NLP.
Key Capabilities
- Structured Medical Note Generation: Converts free-form clinical consultation transcripts into the standardized S, O, A, P format.
- Fine-tuned Performance: Utilizes Supervised Fine-Tuning (SFT) with LoRA, targeting specific language layers for medical text generation.
- Efficient Training: Trained on an H100 GPU using Unsloth and Hugging Face's TRL library, achieving 2x faster training.
- Robust Training Data: Fine-tuned on the
syafiqassegaf/soap-dataset comprising 9,250 examples of medical dialogues and corresponding SOAP notes.
Intended Use & Limitations
This model is designed to assist healthcare professionals and clinical NLP researchers by streamlining the creation of SOAP notes. It is not intended as a substitute for professional medical judgment. Limitations include being trained exclusively on English-language dialogues and potential performance degradation on highly specialized subspecialty consultations not well-represented in its training data. The model converged stably over 5 epochs, with a significant reduction in loss, indicating effective learning of the SOAP output format.