AfriAdapt Agriculture Training Dataset

High-quality instruction-response dataset focused on practical agricultural advice for smallholder farmers in East and Southern Africa.

This dataset was created as part of the AfriAdapt project for the Adaption AutoScientist Challenge 2026.

Dataset Summary

• Domain: Agriculture & Climate-smart farming
• Target users: Smallholder farmers and agricultural extension workers
• Geographic focus: Kenya, Tanzania, Uganda, Ethiopia and similar regions
• Language: English
• Size: 176 high-quality examples
• Format: Instruction → Output pairs

Intended Use

This dataset is designed for:

• Fine-tuning / adapting language models for agricultural advisory tasks
• Improving model performance on practical, resource-constrained farming scenarios
• Research on climate-resilient and locally relevant agricultural AI

Data Fields

Field	Type	Description
instruction	string	Realistic farming scenario / question
output	string	Practical, step-by-step advisory response
domain	string	Always agriculture
language	string	Language of the example (en)
id	string	Unique identifier
metadata	object	Focus area, difficulty, source, quality flag

Example

Instruction:

In the semi-arid region of Machakos County, Kenya, smallholder farmer Jane is experiencing severe drought...

Output:

To manage moisture stress in her maize crop, Jane can take the following steps: first, she should mulch...

Data Creation

The dataset was generated using a carefully designed multi-stage pipeline (AfriAdapt) that emphasizes:

• Realistic smallholder constraints
• Local context (East & Southern Africa)
• Practical and safe recommendations
• Climate resilience and limited-resource decision making

License

Apache 2.0

Citation

@dataset{afriadapt_agri_train_2026,
  title={AfriAdapt Agriculture Training Dataset},
  author={AfriAdapt},
  year={2026},
  url={https://huggingface.co/datasets/vancouverevs/afriadapt-agri-train}
}