Training Dataset for the Name Spell Model

This repo contains five (5) datasets used for training, validating and testing the Name Spell Model hosted on Hugging Face. Each CSV file contains Hausa personal names and their corresponding spelling attempts, which are a combination of real and synthetic data. The experiment involved fine-tuning ByT5 small, Flan-T5 small and AfroLlaMA for the purpose of predicting the correct spelling of misspelt Hausa Personal names. The base and corresponding fine-tuned models were evaluated, and their predictions are provided in the respective CSV files attached herein.