samwell/ncd-gemma4-e4b-lora

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 10, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

The samwell/ncd-gemma4-e4b-lora is a 7.9 billion parameter Gemma 4 E4B model fine-tuned by Samwell for Non-Communicable Disease (NCD) risk assessment. Specifically, it predicts Type 2 Diabetes and Hypertension risk from patient clinical data, optimized for clinical decision support in resource-limited healthcare settings. Trained on 49,214 synthetic patient records, it provides structured risk assessments with clinical reasoning, supporting a 32768 token context length.

Loading preview...

NCD Risk Assessment Model (Gemma 4 E4B Fine-tuned)

This model is a fine-tuned version of Google's gemma-4-E4B-it base model, developed by Samwell, specifically designed for Non-Communicable Disease (NCD) risk assessment. It focuses on predicting the risk of Type 2 Diabetes and Hypertension from patient clinical data.

Key Capabilities & Features

  • Specialized Risk Assessment: Provides structured risk assessments for Type 2 Diabetes and Hypertension based on patient demographics, vitals, and lab values.
  • Clinical Decision Support: Intended to assist healthcare workers in primary care and community health settings, particularly in resource-limited environments like Ghana.
  • Efficient Fine-tuning: Utilizes QLoRA (4-bit) with Unsloth, trained on 39,371 synthetic patient records in approximately 100 minutes on an H200 GPU.
  • Deployment Flexibility: Available as a LoRA adapter for integration with Hugging Face Transformers and PEFT, and as a GGUF quantized model for local inference with llama.cpp or Ollama.

Intended Use Cases

  • NCD Risk Screening: Assisting healthcare professionals in identifying individuals at risk for Type 2 Diabetes and Hypertension.
  • Integration with EHR Systems: Can be integrated into Electronic Health Record (EHR) systems like OpenMRS/HopeOS for automated risk assessment.

Limitations

It's important to note that the model was trained exclusively on synthetic data, not real patient records. It is designed as a screening support tool, not a diagnostic instrument, and requires clinical validation before use in actual patient care. The model currently assesses only diabetes and hypertension.