Clinical LLaMA 3
N/A
GitHub Stars
N/A
Open Issues
N/A
Docker Support
N/A
Last Updated
Technical Summary
Clinical LLaMA 3 is an open-weights Large Language Model built upon the powerful LLaMA 3 architecture, specifically quantized and fine-tuned for structured information extraction from unstructured clinical notes.
Key Capabilities
- Zero-Shot ICD-10 Extraction: Reliably parse complex discharge summaries into discrete, valid, and billable ICD-10 codes without requiring extensive few-shot prompting.
- HIPAA-Compliant Deployment: Can be run entirely locally on commodity hardware using
llama.cppto ensure that no Protected Health Information (PHI) ever leaves the hospital firewall. - Structured JSON Generation: Highly reliable JSON schema adherence, making it a perfect component for automated data pipelines using LangChain or LlamaIndex.
Getting Started
For an end-to-end example of utilizing Clinical LLaMA 3 to automate medical billing, see Cookbook 15: Automating ICD-10 Coding with LLMs in the Getting Started guide.
Model Card Details
Architecture
Transformer-based Large Language Model (8B/70B variants) fine-tuned on clinical text.
Intended Use Cases
Medical billing automation, ICD-10 extraction, unstructured clinical note summarization.
Training Data
MIMIC-III clinical notes, PubMed abstracts, and synthetic EHR records.