PatrickChikuse/chichewa-agri-advisor

一个奇切瓦语指令数据集，用于微调Llama风格模型，为马拉维农民提供农业建议（重点：玉米/chimanga）。数据集包含198个对话，分为训练集和验证集，建议至少500-1000个示例以获得可用的LoRA微调。每个对话遵循OpenAI / Llama聊天消息模式，包含系统提示、用户问题和助手回答。数据集还包括预处理脚本、训练图表和上传到Hugging Face的工具。系统提示为奇切瓦语，限制模型仅回答农业/玉米相关问题。

A Chichewa-language instruction dataset for fine-tuning a Llama-style model to give agricultural advice to Malawian farmers (focus: maize / chimanga). The dataset contains 198 conversations split into training and validation sets, with a recommended minimum of 500–1,000 examples for a usable LoRA finetune. Each conversation follows the OpenAI / Llama chat-message schema, including a system prompt, user question, and assistant answer. The dataset also includes preprocessing scripts, training charts, and tools for uploading to Hugging Face. The system prompt is in Chichewa and constrains the model to answer only agriculture/maize questions.