Qwen2.5-3B ADE Relation Extractor
Standalone merged version of a QLoRA fine-tuned Qwen/Qwen2.5-3B-Instruct model for extracting drug–adverse drug event (ADE) relations from clinical sentences.
Model
- Base model:
Qwen/Qwen2.5-3B-Instruct - Task: ADE relation extraction
- Fine-tuning: QLoRA
- Dataset: ADE Corpus V2
- Train / Validation / Test: 5,570 / 500 / 500
- LoRA:
r=16,alpha=32, dropout=0.05 - Quantization during training: 4-bit NF4
- Compute dtype: BF16
This repository contains the LoRA adapter merged into the base model and can be loaded directly with Transformers.
Results
| Metric | Fine-tuned Model |
|---|---|
| Strict Pair F1 | 71.98% |
| Drug F1 | 92.82% |
| Effect F1 | 76.44% |
| JSON Validity | 99.00% |
| Relaxed Pair F1 | 84.82% |
| G-Eval Average Correctness Score | 0.8749 |
G-Eval was performed with DeepEval using gpt-4o-mini.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "patilshrinivas/Qwen2.5-3B-Instruct-drug-ade-relation-extractor"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
dtype="bfloat16",
device_map="auto",
)
For generation, use the model with the same clinical extraction prompt/schema described in the project documentation.
Intended Use
This model is intended for research and experimentation in structured clinical information extraction.
It is not a medical diagnostic system and does not replace clinical judgement.