BULaMU releases 20M–110M multilingual small models that run offline on phones

On November 13, 2026 at 20:29, the creator of BULaMU shared a post in the Reddit community LocalLLaMA to announce progress on the project. BULaMU is described as a family of language models with 20M, 47M, and 110M parameters, trained completely from scratch with a focus on a low-resource language, Luganda. Over the past several months, the creator says the training corpus was expanded tenfold, reaching 40 billion tokens, with the addition of text from five other East African languages, including Swahili. The models are characterized as small and compute-efficient, designed to run offline on a phone without requiring a GPU or an internet connection, which the creator presents as the core capability of the series. At this stage, the only available information comes from this first-person post by the project's creator; the materials do not provide benchmark results, evaluation figures, download channels, or any follow-up release plan, so the project's current status beyond this announcement cannot be verified from the given reports.