Model

Swahili-gemma-1B-it

A Gemma 3 1B model adapted using LoRA for English to Swahili translation.

  • Swahili
  • Translation
  • Open Model
About this model

Advancing English to Swahili translation

Built on Google's Gemma 3 1B architecture, this model uses LoRA fine-tuning to improve English-to-Swahili translation. It was trained on more than 159,000 examples from MAFAND-MT, Sunbird SALT, Rogendo, Inkuba-Swahili-MMT, Aya, and Alpaca-Swahili, adapting the base model with just 26.09 million trainable parameters merged into full weights.

  • English to Swahili translation
  • Swahili instruction-following and conversational use
  • Low-resource African language NLP research
Hugging Face

Model Information

Language
Swahili
Task
Translation
Parameters
1B · 26.09M trained
Method
Continued pretraining → QLoRA SFT · merged fp16
Direction
en → sw
License
Gemma terms
Released
Aug 11, 2026
Base Model
google/gemma-3-1b-it ↗
Available on
Hugging Face, Ollama, Kaggle
Playground

Try the Model

Explore how the model translates civic discourse from English to Swahili

Swahili-gemma-1B-it

English0/500
Swahili
The translation will appear here