初始化项目，由ModelHub XC社区提供模型
Model: emylton/arogya-ai-full Source: Original Platform
This commit is contained in:
ModelHub XC
364
README.md
Normal file
364
README.md
Normal file
@@ -0,0 +1,364 @@
|
||||
---
|
||||
language:
|
||||
- id
|
||||
- en
|
||||
license: llama3
|
||||
library_name: transformers
|
||||
pipeline_tag: text-generation
|
||||
tags:
|
||||
- healthcare
|
||||
- indonesia
|
||||
- maluku-tenggara
|
||||
- disease-prediction
|
||||
- llama-3
|
||||
- medical-ai
|
||||
base_model: meta-llama/Meta-Llama-3-8B
|
||||
datasets:
|
||||
- custom
|
||||
---
|
||||
|
||||
# Arogya AI - Full Model
|
||||
|
||||
**Arogya** (Sanskrit: "perfect health") adalah model AI kesehatan yang dilatih khusus untuk analisis dan prediksi data kesehatan di Kabupaten Maluku Tenggara, Indonesia.
|
||||
|
||||
🔗 **GitHub Repository**: https://github.com/Rafael2022-prog/arogya-ai
|
||||
|
||||
📄 **Research Paper**:
|
||||
- [English](https://github.com/Rafael2022-prog/arogya-ai/blob/main/PAPER_AROGYA_EN.md)
|
||||
- [Indonesian](https://github.com/Rafael2022-prog/arogya-ai/blob/main/PAPER_AROGYA_ID.md)
|
||||
|
||||
🎯 **LoRA Adapter Version**: [emylton/arogya-health-model](https://huggingface.co/emylton/arogya-health-model) (27 MB)
|
||||
|
||||
---
|
||||
|
||||
## Model Description
|
||||
|
||||
Ini adalah **full merged model** (Llama 3 8B + LoRA adapter) yang siap pakai tanpa perlu download base model terpisah.
|
||||
|
||||
### Key Features
|
||||
|
||||
- **Ready to use**: Tidak perlu base model Llama 3
|
||||
- **Ollama compatible**: Bisa langsung import ke Ollama
|
||||
- **Specialized**: Dilatih dengan 10,000+ data kesehatan real dari Maluku Tenggara
|
||||
- **Multi-disease**: Mendukung 7 penyakit utama (DBD, ISPA, Malaria, Diare, TB, Stunting, Pneumonia)
|
||||
- **Geographic coverage**: 9 kecamatan di Kabupaten Maluku Tenggara
|
||||
|
||||
### Training Data
|
||||
|
||||
Model dilatih menggunakan 10,000+ records dari 4 sumber data:
|
||||
1. **LAMPIRAN PROFIL MALUKU TENGGARA 2023** (Excel)
|
||||
2. **LAMPIRAN PROFIL KESEHATAN MALRA 2024** (Excel)
|
||||
3. **RENJA 2026 DINKES MALRA** (PDF - 24 pages, 14 tables)
|
||||
4. **RENSTRA DINAS KESEHATAN 2025-2029** (PDF - 97 pages, 74 tables)
|
||||
|
||||
Data mencakup 169 indikator kesehatan dari tahun 2020-2029.
|
||||
|
||||
### Model Details
|
||||
|
||||
- **Base Model**: Meta-Llama-3-8B
|
||||
- **Fine-tuning Method**: LoRA (r=16, alpha=32)
|
||||
- **Training**: 3 epochs, batch_size=4, learning_rate=2e-4
|
||||
- **Model Size**: ~16 GB (FP16)
|
||||
- **Parameters**: 8 billion
|
||||
- **Context Length**: 8192 tokens
|
||||
- **Language**: Indonesian & English
|
||||
|
||||
---
|
||||
|
||||
## Quick Start
|
||||
|
||||
### Python (Transformers)
|
||||
|
||||
```python
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
import torch
|
||||
|
||||
# Load model
|
||||
model = AutoModelForCausalLM.from_pretrained(
|
||||
"emylton/arogya-ai-full",
|
||||
torch_dtype=torch.float16,
|
||||
device_map="auto"
|
||||
)
|
||||
|
||||
tokenizer = AutoTokenizer.from_pretrained("emylton/arogya-ai-full")
|
||||
|
||||
# Generate prediction
|
||||
prompt = """Prediksi kasus DBD di Kei Kecil untuk bulan depan berdasarkan data:
|
||||
- Bulan ini: 45 kasus
|
||||
- Bulan lalu: 38 kasus
|
||||
- Curah hujan: tinggi
|
||||
- Musim: penghujan"""
|
||||
|
||||
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
||||
outputs = model.generate(
|
||||
**inputs,
|
||||
max_new_tokens=200,
|
||||
temperature=0.7,
|
||||
top_p=0.9,
|
||||
do_sample=True
|
||||
)
|
||||
|
||||
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
||||
```
|
||||
|
||||
### Ollama
|
||||
|
||||
```bash
|
||||
# 1. Download model
|
||||
huggingface-cli download emylton/arogya-ai-full --local-dir ./arogya-full
|
||||
|
||||
# 2. Create Modelfile
|
||||
cat > Modelfile << EOF
|
||||
FROM ./arogya-full
|
||||
PARAMETER temperature 0.7
|
||||
PARAMETER top_p 0.9
|
||||
PARAMETER num_ctx 8192
|
||||
SYSTEM "Saya Arogya AI, asisten kesehatan untuk Kabupaten Maluku Tenggara. Saya dapat membantu analisis data kesehatan, prediksi penyakit, dan rekomendasi intervensi untuk 7 penyakit utama: DBD, ISPA, Malaria, Diare, TB, Stunting, dan Pneumonia."
|
||||
EOF
|
||||
|
||||
# 3. Import to Ollama
|
||||
ollama create arogya-ai -f Modelfile
|
||||
|
||||
# 4. Run
|
||||
ollama run arogya-ai "Prediksi kasus DBD di Kei Kecil"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Use Cases
|
||||
|
||||
### 1. Disease Prediction
|
||||
```python
|
||||
prompt = "Prediksi kasus Malaria di Kei Besar bulan Maret 2026"
|
||||
```
|
||||
|
||||
### 2. Trend Analysis
|
||||
```python
|
||||
prompt = "Analisis trend kasus ISPA di Maluku Tenggara 2023-2024"
|
||||
```
|
||||
|
||||
### 3. Intervention Recommendations
|
||||
```python
|
||||
prompt = "Rekomendasi intervensi untuk menurunkan kasus Stunting di Kei Kecil"
|
||||
```
|
||||
|
||||
### 4. Resource Allocation
|
||||
```python
|
||||
prompt = "Alokasi sumber daya kesehatan untuk program TB di 9 kecamatan"
|
||||
```
|
||||
|
||||
### 5. Risk Assessment
|
||||
```python
|
||||
prompt = "Penilaian risiko outbreak DBD di musim hujan"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Supported Diseases
|
||||
|
||||
1. **DBD** (Demam Berdarah Dengue)
|
||||
2. **ISPA** (Infeksi Saluran Pernapasan Akut)
|
||||
3. **Malaria**
|
||||
4. **Diare**
|
||||
5. **TB** (Tuberkulosis)
|
||||
6. **Stunting**
|
||||
7. **Pneumonia**
|
||||
|
||||
## Geographic Coverage
|
||||
|
||||
9 Kecamatan di Kabupaten Maluku Tenggara:
|
||||
- Kei Kecil
|
||||
- Kei Besar
|
||||
- Kei Besar Selatan
|
||||
- Kei Besar Utara Timur
|
||||
- Kei Besar Utara Barat
|
||||
- Hoat Sorbay
|
||||
- Manyeuw
|
||||
- Kei Kecil Timur
|
||||
- Kei Kecil Barat
|
||||
|
||||
---
|
||||
|
||||
## Training Details
|
||||
|
||||
### Hyperparameters
|
||||
|
||||
```python
|
||||
{
|
||||
"lora_r": 16,
|
||||
"lora_alpha": 32,
|
||||
"lora_dropout": 0.05,
|
||||
"learning_rate": 2e-4,
|
||||
"num_train_epochs": 3,
|
||||
"per_device_train_batch_size": 4,
|
||||
"gradient_accumulation_steps": 4,
|
||||
"warmup_steps": 100,
|
||||
"max_seq_length": 2048,
|
||||
"optimizer": "paged_adamw_8bit"
|
||||
}
|
||||
```
|
||||
|
||||
### Training Infrastructure
|
||||
|
||||
- **Platform**: Google Colab Pro
|
||||
- **GPU**: NVIDIA A100 (40GB)
|
||||
- **Training Time**: ~6 hours
|
||||
- **Framework**: Transformers + PEFT + bitsandbytes
|
||||
|
||||
### Data Processing
|
||||
|
||||
Data diproses melalui pipeline:
|
||||
1. **Excel Extraction**: Pandas untuk structured data
|
||||
2. **PDF Extraction**: PyMuPDF + tabula untuk tables
|
||||
3. **Data Cleaning**: Normalisasi, deduplication
|
||||
4. **Prompt Engineering**: Template khusus untuk health data
|
||||
5. **Train/Val Split**: 90/10
|
||||
|
||||
---
|
||||
|
||||
## Limitations
|
||||
|
||||
⚠️ **Important Limitations:**
|
||||
|
||||
1. **Geographic Specificity**: Model dilatih khusus untuk Maluku Tenggara, mungkin kurang akurat untuk daerah lain
|
||||
2. **Disease Coverage**: Hanya 7 penyakit utama, tidak mencakup semua kondisi kesehatan
|
||||
3. **Data Timeframe**: Data training dari 2020-2024, prediksi jangka panjang mungkin kurang akurat
|
||||
4. **Not Medical Advice**: Model ini untuk analisis data, bukan pengganti konsultasi medis profesional
|
||||
5. **Language**: Optimal untuk Bahasa Indonesia, kemampuan bahasa lain terbatas
|
||||
6. **Hallucination**: Seperti LLM lainnya, model dapat menghasilkan informasi yang tidak akurat
|
||||
|
||||
---
|
||||
|
||||
## Ethical Considerations
|
||||
|
||||
### Intended Use
|
||||
|
||||
✅ **Recommended:**
|
||||
- Analisis data kesehatan populasi
|
||||
- Perencanaan program kesehatan
|
||||
- Alokasi sumber daya
|
||||
- Penelitian epidemiologi
|
||||
- Edukasi kesehatan masyarakat
|
||||
|
||||
❌ **Not Recommended:**
|
||||
- Diagnosis medis individual
|
||||
- Keputusan klinis tanpa verifikasi profesional
|
||||
- Pengganti tenaga kesehatan
|
||||
- Situasi darurat medis
|
||||
|
||||
### Privacy & Security
|
||||
|
||||
- Model tidak menyimpan data personal
|
||||
- Tidak ada identitas pasien dalam training data
|
||||
- Semua data diagregasi di level populasi
|
||||
- Ikuti regulasi kesehatan lokal saat menggunakan
|
||||
|
||||
---
|
||||
|
||||
## Performance
|
||||
|
||||
### Evaluation Metrics
|
||||
|
||||
Model dievaluasi pada validation set (10% dari data):
|
||||
|
||||
- **Perplexity**: 2.34
|
||||
- **Loss**: 0.85
|
||||
- **Accuracy** (classification tasks): 87.3%
|
||||
- **F1 Score**: 0.86
|
||||
|
||||
### Comparison
|
||||
|
||||
| Model | Size | Accuracy | Use Case |
|
||||
|-------|------|----------|----------|
|
||||
| Base Llama 3 8B | 16 GB | 45.2% | General |
|
||||
| Arogya (LoRA) | 27 MB | 87.3% | Health (need base) |
|
||||
| Arogya (Full) | 16 GB | 87.3% | Health (standalone) |
|
||||
|
||||
---
|
||||
|
||||
## Model Versions
|
||||
|
||||
### Full Model vs LoRA Adapter
|
||||
|
||||
| Aspect | LoRA Adapter | Full Model |
|
||||
|--------|--------------|------------|
|
||||
| **Repository** | [emylton/arogya-health-model](https://huggingface.co/emylton/arogya-health-model) | [emylton/arogya-ai-full](https://huggingface.co/emylton/arogya-ai-full) |
|
||||
| **Size** | 27 MB | ~16 GB |
|
||||
| **Download Time** | 1 min | 10-30 min |
|
||||
| **Requires Base Model** | ✅ Yes (Llama 3 8B) | ❌ No |
|
||||
| **Ollama Compatible** | ❌ No | ✅ Yes |
|
||||
| **Best For** | Developers/Researchers | End Users |
|
||||
|
||||
**Recommendation**:
|
||||
- Use **LoRA adapter** if you already have Llama 3 8B or want to experiment
|
||||
- Use **Full model** for production deployment or Ollama usage
|
||||
|
||||
---
|
||||
|
||||
## Citation
|
||||
|
||||
If you use this model in your research, please cite:
|
||||
|
||||
```bibtex
|
||||
@software{arogya_ai_2024,
|
||||
title = {Arogya AI: Fine-tuned Language Model for Health Data Analysis in Maluku Tenggara},
|
||||
author = {Rafael and Contributors},
|
||||
year = {2024},
|
||||
url = {https://huggingface.co/emylton/arogya-ai-full},
|
||||
note = {Based on Meta-Llama-3-8B}
|
||||
}
|
||||
```
|
||||
|
||||
**Research Paper**:
|
||||
```bibtex
|
||||
@article{arogya_paper_2024,
|
||||
title = {Arogya AI: Implementasi Large Language Model untuk Analisis dan Prediksi Data Kesehatan di Kabupaten Maluku Tenggara},
|
||||
author = {Rafael and Contributors},
|
||||
year = {2024},
|
||||
url = {https://github.com/Rafael2022-prog/arogya-ai}
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## License
|
||||
|
||||
This model is based on Meta-Llama-3-8B and follows the [Llama 3 Community License](https://llama.meta.com/llama3/license/).
|
||||
|
||||
**Additional Terms:**
|
||||
- Model dapat digunakan untuk tujuan penelitian dan komersial
|
||||
- Wajib mencantumkan attribution
|
||||
- Tidak untuk tujuan yang merugikan atau melanggar hukum
|
||||
- Ikuti regulasi kesehatan setempat
|
||||
|
||||
---
|
||||
|
||||
## Acknowledgments
|
||||
|
||||
- **Meta AI** untuk Llama 3 base model
|
||||
- **Dinas Kesehatan Kabupaten Maluku Tenggara** untuk data kesehatan
|
||||
- **Hugging Face** untuk platform dan tools
|
||||
- **Google Colab** untuk training infrastructure
|
||||
|
||||
---
|
||||
|
||||
## Contact & Support
|
||||
|
||||
- **GitHub Issues**: https://github.com/Rafael2022-prog/arogya-ai/issues
|
||||
- **Model Repository**: https://huggingface.co/emylton/arogya-ai-full
|
||||
- **Documentation**: https://github.com/Rafael2022-prog/arogya-ai
|
||||
|
||||
---
|
||||
|
||||
## Updates
|
||||
|
||||
### Version 1.0 (2024)
|
||||
- Initial release
|
||||
- 10,000+ training samples
|
||||
- 7 diseases coverage
|
||||
- 9 sub-districts coverage
|
||||
- Full model deployment
|
||||
|
||||
---
|
||||
|
||||
**Built with ❤️ for better healthcare in Maluku Tenggara**
|
||||
Reference in New Issue
Block a user