MichaelMukiibi/yogera

Serving Ganda Gemma 1B inference from the browser # Inference with Ganda_Gemma 1B LiteRT + MediaPipe Web Serving an LLM via the browser using MediaPipe Web. Gemma 1B series is series of machine learning models released by Google to enable edge AI inference. Crane AI Labs, based in Kampala, Uganda finetuned this model to speak Luganda, the most widely spoken language in Uganda. Using UnSloth QLoRA for finetuning achieving higher scores on BLEU compared to Gemma 27B.