|
|
--- |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- agriculture |
|
|
- farming |
|
|
- qa |
|
|
- lora |
|
|
- peft |
|
|
- qwen |
|
|
license: mit |
|
|
datasets: |
|
|
- shchoi83/agriQA |
|
|
base_model: Qwen/Qwen1.5-1.8B-Chat |
|
|
--- |
|
|
|
|
|
# 🌾 AgriQA Assistant |
|
|
|
|
|
An intelligent agricultural expert assistant fine-tuned on the agriQA dataset using Qwen1.5-1.8B-Chat with PEFT + LoRA. |
|
|
|
|
|
## 🚀 Features |
|
|
|
|
|
- **Clear, practical steps** you can apply directly in the field |
|
|
- **Specific measurements and quantities** for accurate application |
|
|
- **Safety precautions** when needed |
|
|
- **Expert tips** for better results |
|
|
- **Structured responses** with numbered steps |
|
|
|
|
|
## 🔧 Technical Details |
|
|
|
|
|
- **Base Model**: Qwen/Qwen1.5-1.8B-Chat |
|
|
- **Fine-tuning Method**: PEFT + LoRA (Parameter Efficient Fine-tuning) |
|
|
- **Dataset**: agriQA (agricultural Q&A pairs) |
|
|
- **Training Data**: 50,000 samples with structured prompts |
|
|
- **LoRA Rank**: 2 |
|
|
- **LoRA Alpha**: 4 |
|
|
|
|
|
## 📱 Usage |
|
|
|
|
|
### Direct Usage |
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
# Load base model |
|
|
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen1.5-1.8B-Chat", trust_remote_code=True) |
|
|
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-1.8B-Chat", trust_remote_code=True) |
|
|
|
|
|
# Load LoRA adapter |
|
|
model = PeftModel.from_pretrained(base_model, "nada013/agriqa-assistant") |
|
|
``` |
|
|
|
|
|
### Chat Format |
|
|
```python |
|
|
messages = [ |
|
|
{"role": "system", "content": "You are AgriQA, an agricultural expert assistant..."}, |
|
|
{"role": "user", "content": "How to control aphid infestation in mustard crops?"} |
|
|
] |
|
|
|
|
|
# Generate response |
|
|
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt") |
|
|
outputs = model.generate(inputs, max_new_tokens=512, temperature=0.3) |
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
``` |
|
|
|
|
|
## 🎯 Response Format |
|
|
|
|
|
The model provides structured responses: |
|
|
1. **Direct answer** to the question |
|
|
2. **Numbered step-by-step solution** |
|
|
3. **Specific details** (measurements, quantities, product names) |
|
|
4. **Safety precautions** if needed |
|
|
5. **Extra tip or follow-up advice** |
|
|
|
|
|
## 💡 Example Questions |
|
|
|
|
|
- "How to control aphid infestation in mustard crops?" |
|
|
- "What fertilizer should I use for coconut plants?" |
|
|
- "How to increase milk production in cows?" |
|
|
- "What is the treatment for white diarrhoea in poultry?" |
|
|
- "How to preserve potato tubers for 7-8 months?" |
|
|
|
|
|
## 🔒 Safety Note |
|
|
|
|
|
Always follow safety guidelines when applying agricultural practices. The assistant provides general advice - consult local agricultural experts for region-specific recommendations. |
|
|
|
|
|
## 📊 Training Details |
|
|
|
|
|
- **Epochs**: 1 |
|
|
- **Learning Rate**: 5e-4 |
|
|
- **Batch Size**: 1 (with gradient accumulation) |
|
|
- **Max Length**: 256 tokens |
|
|
- **Optimizer**: AdamW with fused implementation |
|
|
- **Hardware**: 8GB GPU with 4-bit quantization |
|
|
|
|
|
## 🤝 Contributing |
|
|
|
|
|
This model is trained on the agriQA dataset. For improvements or questions, please refer to the original dataset source. |
|
|
|
|
|
## 📄 License |
|
|
|
|
|
This project uses the Qwen1.5-1.8B-Chat model and agriQA dataset. Please refer to their respective licenses for usage terms. |
|
|
|
|
|
--- |
|
|
|
|
|
**Built with ❤️ for the agricultural community** |
