PatrickChikuse/chichewa-agriculture-advisory

Chichewa农业咨询数据集是一个用于微调Llama风格聊天模型的Chichewa语指令数据集，专注于为马拉维农民提供玉米种植建议。数据集包含198行对话数据，分为训练集（178行）和验证集（20行），采用90/10的比例分割。每条数据遵循OpenAI/Llama聊天消息模式，包含系统消息、用户问题和助手回答。系统消息统一为农业/玉米主题的提示。数据集主要用于微调小型Llama风格模型（LoRA/QLoRA）进行Chichewa农业问答，以及测试低资源非洲语言的聊天流程。数据集的局限性包括数据量小、作物覆盖不均衡、单一系统提示以及未经专业审核。

A Chichewa-language instruction dataset for fine-tuning a Llama-style chat model to advise Malawian farmers, with a focus on maize (*chimanga*). The dataset contains 198 rows of conversation data, split into a training set (178 rows) and a validation set (20 rows) with a 90/10 split ratio. Each data follows the OpenAI/Llama chat-message schema, including system message, user question, and assistant answer. The system message is uniformly set to agriculture/maize topics. The dataset is mainly used for fine-tuning small Llama-style models (LoRA/QLoRA) for Chichewa farming Q&A and smoke-testing chat pipelines for low-resource African languages. Limitations include small size, uneven crop coverage, single system prompt, and lack of professional review.