Nahara Dataset Model
- Developed by: Redeemer Salami Okekale, BMS
- License: apache-2.0
- Finetuned from model: unsloth/meta-llama-3.1-8b-bnb-4bit
- Training Loss: 1.181600
Model Description:
The nahara-dataset-model is a fine-tuned version of Meta's LLaMA series, specifically optimized for low-precision (4-bit) operations to enhance efficiency in both memory usage and computational resources. It was fine-tuned on the Nahara dataset and achieved a training loss of 1.181600, ensuring strong performance on medical data.
- Model Type: Transformer-based Language Model
- Size: 8 billion parameters
- Precision: 4-bit quantization using bnb (bits and bytes), improving memory efficiency and making the model suitable for resource-constrained environments.
Intended Use:
This model serves as a highly adaptable AI copilot for medical professionals, ideal for providing real-time recommendations and decision support. It can assist with:
- Medical diagnostics and treatment suggestions
- Summarization of clinical data
- Generation of medical reports and documentation
- Assistance with medical coding and research data preparation
Performance:
- Training Loss: 1.181600
- Fine-tuning Data: Medical and clinical datasets enhanced through data augmentation techniques to handle sparsity and variability, making it applicable across various healthcare contexts.
Applications:
The nahara-dataset-model is suited for:
- Clinical decision support systems
- AI copilots for medical professionals
- Research data analysis and augmentation
- Medical record summarization and automated report generation
Limitations and Considerations:
- The model is trained on medical data but may not encompass all clinical expertise nuances. It should be used to augment decision-making, not replace professional judgment.
- Ethical considerations, including data privacy and bias in healthcare applications, must be strictly followed.
- While efficiency is boosted by quantizing to 4-bit, there may be trade-offs in performance for complex tasks compared to higher precision models.
Future Improvements:
The model will undergo further optimization and refinement in Phase 2, including expanding the dataset, improving real-world adaptability, and fine-tuning the AI copilot for specific medical specializations.
Contributors:
- Emmanuel Akomanin Asiamah, PhD
- Elli Banini
- Felix Coker
- Philip Attram, BMS
- Schandorf Osam-Frimpong, MD
- Daniel Mawuenyega Gohoho
- Vitus Amenorpe
- Aaron Kofi Gayi
- Julius Richard Ogbey
- Cherryln Asiwome Ahiable
- Ama Quashie
- Andrew Kojo Mensah-Onumah
- Edith Zikpi
- Azumah Benson, MD
