LoRA fine-tune of microsoft/phi-1_5 on the Kaggle Crop Recommendation Dataset (2,200 real records: soil N-P-K, pH, temperature, humidity, rainfall -> recommended crop).

Given soil nutrient and climate readings, recommends one of 22 crops.

General farming advice, pest/disease guidance, yield predictions, or anything outside the soil->crop mapping it was trained on. It will produce plausible-sounding but fabricated numbers (e.g. yield figures) for anything outside its training data -- do not trust specific numeric claims from it.

Prompt format: Question: <soil/climate description>\nAnswer:

Kaggle Crop Recommendation Dataset by Atharva Ingle -- check the dataset's license before assuming redistribution rights for the raw data itself.