johnemekaeze/LowResource-Text2SQL-Chichewa

Here, we focus on developing strategies to Improving LLM-based Text-to-SQL systems for Low-Resource Languages - A Case study of Chichewa # Adapting LLMs for Text-to-SQL Generation in Low-Resource Languages: A Case Study of Chichewa ## Master's Thesis by John Emeka Eze (AIMS Rwanda, 2025) ### Overview This research addresses the gap in Text-to-SQL systems for low-resource languages (LRLs) by: - Proposing a cross-lingual framework combining QLoRA fine-tuning, prompt engineering, and retrieval augmentated semantic parsing. - Creating the small-scale Chichewa NL-SQL benchmark dataset - Evaluating adaptation strategies for LRLs using Chichewa (spoken by 12M+ people) ### Note - `data` folder contains a small-scale Chichewa benchmark. It is important to note that the Spider 1.0 dataset alongside the provided data as shown in the notebooks. - `docs` folder contains the complete thesis and tables of results - `notebooks` folder contains codes of all the experiments carried out #### Ignore the `metadata.widgets` error displayed when you click the notebook. Download them to have access to the codes.