agribot-lora

🌱 AgriBot — A Fine-tuned Agriculture Assistant (LoRA)

AgriBot is a lightweight language model fine-tuned to assist farmers and agri-stakeholders with agriculture-related queries using LoRA (Low-Rank Adaptation). It is based on the distilgpt2 model and trained on a curated dataset of agricultural Q&A to deliver relevant, natural-language answers.

🔍 Model Details

• Base Model: distilgpt2
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Dataset Used: Keshav022/Agriculture-Dataset
• Training Platform: Google Colab with 8-bit quantization (bitsandbytes)
• Frameworks: Hugging Face transformers, peft, datasets

---

📚 Dataset Summary

The dataset contains real-world agriculture-related questions and answers including topics such as:

• Crop selection and yield optimization
• Soil nutrient management
• Fertilizer and pesticide usage
• Plant disease control
• Seasonal farming practices

We used a cleaned and shuffled subset (~10,000 examples) for efficient training within limited compute resources.

---

🧠 Training Configuration

---

🗣️ Example Usage

from transformers import pipeline

pipe = pipeline("text-generation", model="subbuks/agribot-lora")

output = pipe("Q: How to increase tomato yield?\nA:", 
              max_new_tokens=50, 
              temperature=0.7, 
              top_p=0.95,
              repetition_penalty=1.2,
              do_sample=True)

print(output[0]['generated_text'])

💡 Sample Output

Input:

Q: How to increase tomato yield? A:

Output:

A: Apply organic compost, maintain proper spacing, and water regularly. Use phosphorus-rich fertilizer during flowering.

⚠️ Limitations

Not suitable for high-stakes decision-making without expert verification.

Might hallucinate or repeat answers if generation parameters are not well-tuned.

Trained on English data; multilingual support is not included.

👤 Author

Created by [subbuks] Model hosted on Hugging Face at: https://huggingface.co/subbuks/agribot-lora