A Chichewa-language instruction dataset for fine-tuning a Llama-style model
to give agricultural advice to Malawian farmers (focus: maize / chimanga).
Total conversations: 198
Train / val split: 178 / 20 (90/10, seed=42)
Recommended minimum: ~500–1,000 examples for a usable LoRA finetune.
198 is enough to smoke-test the pipeline; expect underfitting on real prompts.
Folder layout