|
|
---
|
|
|
license: mit
|
|
|
datasets:
|
|
|
- sajjadhadi/disease-diagnosis-dataset
|
|
|
base_model:
|
|
|
- Qwen/Qwen2.5-3B
|
|
|
pipeline_tag: text-classification
|
|
|
tags:
|
|
|
- biology
|
|
|
language:
|
|
|
- zho
|
|
|
- eng
|
|
|
- fra
|
|
|
- spa
|
|
|
- por
|
|
|
- deu
|
|
|
- ita
|
|
|
- rus
|
|
|
- jpn
|
|
|
- kor
|
|
|
- vie
|
|
|
- tha
|
|
|
- ara
|
|
|
library_name: adapter-transformers
|
|
|
---
|
|
|
|
|
|
# Disease Diagnosis Adapter
|
|
|
|
|
|
A fine-tuned adapter for the Qwen/Qwen2.5-3B model specialized in disease diagnosis and classification.
|
|
|
Trained through MLX and MPI, to test performance and accuracy.
|
|
|
|
|
|
## Overview
|
|
|
|
|
|
This adapter enhances the base Ministral-3b-instruct model to improve performance on medical diagnosis tasks. It was trained on the [disease-diagnosis-dataset](https://huggingface.co/datasets/sajjadhadi/disease-diagnosis-dataset).
|
|
|
The data is over-saturated in some diagnosis, I limit the number of diagnosis and take a limit number of them as training tags.
|
|
|
|
|
|
|
|
|
## Usage
|
|
|
|
|
|
```python
|
|
|
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
|
|
|
|
|
# Load model and tokenizer
|
|
|
model_name = "naifenn/diagnosis-adapter"
|
|
|
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
|
|
# Example input
|
|
|
text = "Patient presents with fever, cough, and fatigue for 3 days."
|
|
|
inputs = tokenizer(text, return_tensors="pt")
|
|
|
|
|
|
# Get prediction
|
|
|
outputs = model(**inputs)
|
|
|
prediction = outputs.logits.argmax(-1).item()
|
|
|
print(f"Predicted diagnosis: {model.config.id2label[prediction]}") |
