Training, evaluation, and diagnostic code for parameter-efficient fine-tuning of multimodal large la
Builds a clean Luganda–English paired speech dataset with translation, TTS synthesis, validation, an