The-African-Research-Collective/translation_bench

# Translation Experiments A suite of translation experiments across a variety of models and datasets, focused on African languages. We believe synthetic data is a powerful tool for improving model performance on low-resource languages, and machine translation is a critical part of that pipeline. We hope these results will help the community understand the strengths and weaknesses of different models and approaches to translation. ## Ablations - Baseline results on the following datasets: - AfriDocMT (Health, Tech domain) - openlanguagedata/flores_plus - SFT with different model and dataset mixture combinations - GRPO on base instruction-tuned models and SFT models - SDPO on base instruction-tuned models and SFT models