BULaMU v3
This is the repository for the finetuned ("story-telling") model of the third version of BULaMU. It has been trained from scratch on 10 billion tokens which includes text 6 East African Languages (Swahili, Luganda, Luo, Amharic, Somali, Kinyaranda) and English. East African Languages make up nearly 90% of the new training corpus. For the most up to date information on BULaMU please follow my X account.
BULaMU v3 currently comes in one version 42M parameters. This family of tiny language models was trained using modified scripts from Andrej Karpathy's llama2.c repo.
For more technical details about BULaMU, please refer to the white paper.
Live Interactive Demo
Try it here.