TranslatePsy-AfriSLM outperforms Gemma-27B on African languages
August 20, 2026
The TranslatePsy-AfriSLM framework uses quality-estimation filtering to remove 96% of training tokens, prioritizing high-quality synthetic data. The resulting fine-tuned small language models outperform much larger systems like TranslateGemma-27B across 19 Sub-Saharan languages.
HOW THIS AFFECTS YOU
●
builderYou can deploy highly efficient SLMs for multilingual applications in African markets.
●
researcherThe study demonstrates the efficiency of curated synthetic data in the quality-efficiency Pareto frontier.