xuzhang0112/Llama-3.1-8B-SCL-ICD10

Llama3.1-8B-SCL-ICD10-LoRA

This repository contains LoRA adapter weights for evidence-based ICD-10 coding, trained on top of Llama-3.1-8B-Instruct. This model first extracts evidence from clinical text and then assigns ICD-10 codes.

We train the model via SCL (Span-Centric Learning) in the paper From Documents to Spans: Scalable Supervision for Evidence-Based ICD Coding with LLMs.

Before inference, please download the base model Llama-3.1-8B-Instruct and then either:

1. 1.merge the LoRA adapter into the base model for deployment, or
2. 2.load the adapter with PEFT at runtime.

During inference, please use the following prompt:

Task:
You are a clinical coding assistant. 

Your task is to analyze the provided clinical note, 
first extract all relevant clinical evidences that supports diagnostic coding, 
and then output the corresponding ICD-10-CM codes.

Example

### Clinical Note:
...

### Evidence

CAD
COPD
Anemia

### ICD-10-CM Codes

I25.10 - Atherosclerotic heart disease of native coronary artery without angina pectoris
J44.9 - Chronic obstructive pulmonary disease, unspecified
D62 - Acute posthemorrhagic anemia

---

### Clinical Note:
{text}

Tips

1. 1.The model outputs free-form text. Please implement your own parsing function to extract structured fields.
2. 2.In our paper, we filter the predicted codes to test set label space to campre it with discriminative models (PLM-ICD); However, we recommend full code system evaluation because it is more aligned with real-world deployment needs. In this setting, our model significantly outpefroms LLMs without fine-tuning.