Hugging Face Overview
StanfordAIMI/GREEN-RadLlama2-7b is a specialized 7 billion parameter language model developed by StanfordAIMI. It is a fine-tuned iteration of the RadLLaMA-7b model, specifically adapted for applications within the medical domain. The model's primary function is to assess and quantify the discrepancies between a given reference radiology report and a candidate radiology report, particularly for Chest X-ray examinations.
Key Capabilities
- Radiology Report Comparison: Designed to evaluate differences between two radiology reports.
- Chest X-ray Focus: Optimized for analyzing reports related to Chest X-ray imaging.
- Fine-tuned Performance: Achieved a final validation loss of 0.0644 during training, indicating strong performance on its specific task.
Training Details
The model was trained with a learning rate of 0.0001 over 3 epochs, utilizing a total batch size of 2048 across 8 GPUs. The training process involved Adam optimizer with cosine learning rate scheduling and a warmup ratio of 0.05. Further details are available on the project website: https://stanford-aimi.github.io/green.html.
Intended Use Cases
This model is intended for use in scenarios requiring automated comparison and evaluation of radiology reports, specifically for Chest X-rays. It can assist in quality control, report generation validation, or educational tools within a medical context.