33 lines
1020 B
Markdown
33 lines
1020 B
Markdown
|
---
|
||
|
|
license: mit
|
||
|
|
base_model: microsoft/phi-1_5
|
||
|
|
tags:
|
||
|
|
- agriculture
|
||
|
|
- india
|
||
|
|
- lora
|
||
|
|
- crop-recommendation
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
library_name: transformers
|
||
|
|
---
|
||
|
|
|
||
|
|
# Agri-SLM India — Crop Recommendation
|
||
|
|
|
||
|
|
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).
|
||
|
|
|
||
|
|
## What it does
|
||
|
|
Given soil nutrient and climate readings, recommends one of 22 crops.
|
||
|
|
|
||
|
|
## What it does NOT do
|
||
|
|
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.
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
Prompt format: `Question: <soil/climate description>\nAnswer:`
|
||
|
|
|
||
|
|
## Training data
|
||
|
|
Kaggle Crop Recommendation Dataset by Atharva Ingle -- check the dataset's license
|
||
|
|
before assuming redistribution rights for the raw data itself.
|