Aaayushiii/mt5-crop-lora

🌱 mT5 Crop Recommendation (LoRA Fine-tuned)

This is a fine-tuned mT5 model using LoRA adapters for crop recommendation tasks.
It takes weather and environmental inputs and suggests the most suitable crop(s) along with profitability insights.

🧑‍🏫 Model Details

• Base Model: google/mt5-base
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Framework: 🤗 Transformers
• Dataset: Custom crop recommendation dataset (weather, soil, profitability annotations)
• Languages: English

📊 Training

• Epochs: ~0.1–1 (early stop around loss 0.1)
• Trainable Parameters: ~344K (LoRA only)
• Total Parameters: ~300M
• Hardware: A100 GPU (Colab)

🔎 Example Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "your-username/mt5-crop-lora"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

query = "Suggest best crop given: rainfall=200mm, temperature=25C, soil=loamy"

inputs = tokenizer(query, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Sample Output:

Best crop: Gram.  
Top 3: Gram, Mustard, Wheat.  
Weather outlook: Cooler, dry weather with lower rainfall.  
Profitability: Moderate.

✅ Use Cases

• Agricultural planning
• Crop advisory chatbots
• Climate-aware farming assistance

⚠️ Limitations

• Limited dataset → may not generalize globally
• Should not replace expert advice
• Cross-check with local agronomic data

📜 License

Apache 2.0

🙌 Acknowledgements

• Google Research for mT5
• Hugging Face Transformers & PEFT