odin-llama3.1-medical-ner-v14

A LoRA adapter fine-tuned for medical Named Entity Recognition (NER) and Relation Extraction (RE).

Task

Extracts medical entities and their relationships from clinical text.

• Entity types: Disease, Drug, Symptom
• Relation types: associated_with, causes, interacts_with, treats

Model Details

• Base model: unsloth/meta-llama-3.1-8b-bnb-4bit
• LoRA rank (r): 32
• LoRA alpha: 32
• Target modules: q_proj, gate_proj, v_proj, up_proj, down_proj, o_proj, k_proj
• PEFT type: LORA

Evaluation Results

Metric	Precision	Recall	F1
Entity (micro)	0.900	0.923	0.911
Relation (micro)	0.814	0.851	0.832

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = "unsloth/meta-llama-3.1-8b-bnb-4bit"
adapter_id = "pabloformoso/odin-llama3.1-medical-ner-v14"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
model = AutoModelForCausalLM.from_pretrained(base_model, device_map="auto")
model = PeftModel.from_pretrained(model, adapter_id)

prompt = """### Instruction:
Extract all medical entities and their relations from the following clinical text.

### Input:
The patient developed acute renal failure after treatment with enalapril.

### Output:"""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Data

Combined dataset from:

• ADE Corpus V2: Drug–adverse effect relations
• BC5CDR: Chemical–disease relations
• BioRED: Biomedical relation extraction