biomistral-cardiac-acsm

Model Card: BioMistral-Cardiac-ACSMOverviewDeveloper: Bilal JaouadBase Model: BioMistral/BioMistral-7B-BnB.4Domain: Cardiovascular rehabilitation and ACSM clinical exercise prescription.Methodology: Domain adaptation (continued pre-training) via QLoRA (4-bit NF4 quantization, LoRA rank $r=16$) on an NVIDIA T4 GPU.Training DatasetThe model was adapted on a rigorously cleaned and curated corpus comprising five core reference documents:Patient Workbook: A 120-page cardiac rehabilitation patient education manual (with blank progress logs filtered out).Clinical Guidelines: The 146-page CPG STEMI 2019 guidelines, utilizing Camelot for structured table extraction.Clinical Protocols: The Malaysian CR-MTAC Protocol 2020 (with non-English annexes safely quarantined).International Standards: The WHO Package of Interventions for Rehabilitation, processed via semantic chunking.Prevention Updates: The ABCs of Primary Cardiovascular Prevention 2019, integrating OCR fallback for text embedded in vector diagrams.Key Features & CapabilitiesClinical Knowledge Internalization: Successfully ingests complex cardiac protocols, patient education workbooks, and clinical practice guidelines.Safe Exercise Classification: Designed to evaluate patient clinical vignettes and assist in determining safe exercise intensities (Light, Moderate, Vigorous) based on standardized physiological parameters (METs, RPE, heart rate).Improved Safety Boundaries: Demonstrates superior clinical discrimination compared to generalist open models, avoiding extreme over-prescription errors.Intended Use & LimitationsIntended Use: Academic research, clinical AI benchmarking, and exploring the feasibility of open-source models in supporting cardiac patients.Important Disclaimer: This model is not a medical device. It is a domain-adapted base model rather than an instruction-aligned system (SFT), meaning it may require few-shot prompting and should never be used for autonomous medical diagnosis or prescription without direct professional supervision.