Introduction

During our Natural Language Processing (NLP) studies, I developed a Kinyarwanda-to-English translation application using Python and Streamlit. The purpose of my project is to help users translate Kinyarwanda sentences into English.

While testing my application, I noticed that my small model did not always produce accurate translations. Sometimes, it generated English sentences that did not preserve the meaning of the original Kinyarwanda sentence. This showed me that building a translation application is not only about writing code. Choosing the right language model is also important.

In this blog, I will research three open-weight Large Language Models (LLMs): Qwen2.5-1.5B-Instruct, Llama 3.2 3B Instruct, and Gemma 3 4B Instruct. I will compare their features and choose the one I want to test first in my project.

1. Qwen2.5-1.5B-Instruct

Qwen2.5-1.5B-Instruct was developed by the Qwen team. It is an instruction-tuned language model with approximately 1.54 billion parameters.…