This study evaluates the performance of Transformer models in translating the low-resource English-Irish language pair. The research demonstrates that appropriate hyperparameter optimization significantly improves translation performance. The choice of subword model is identified as the most critical factor, with SentencePiece models using both unigram and BPE approaches being tested. The study used a 55k DGT corpus and an 88k public admin corpus for evaluation. A Transformer model with a 16k BPE subword model achieved a BLEU score improvement of 7.8 points compared to a baseline RNN model. The study also shows that Transformer models outperform RNN models in translation accuracy, with the best Transformer model achieving a BLEU score of 60.5 and a TER score of 0.33. The research highlights the importance of subword model selection and hyperparameter optimization in low-resource language translation. The study also addresses the environmental impact of model development, noting that energy consumption during model training was tracked. The results indicate that using a 16k BPE subword model in a Transformer architecture leads to significant performance improvements in low-resource English-Irish translation. The study concludes that hyperparameter optimization and subword model selection are crucial for achieving high performance in low-resource language translation. The research also shows that the Transformer model outperforms Google Translate in translation accuracy for the English-Irish language pair.This study evaluates the performance of Transformer models in translating the low-resource English-Irish language pair. The research demonstrates that appropriate hyperparameter optimization significantly improves translation performance. The choice of subword model is identified as the most critical factor, with SentencePiece models using both unigram and BPE approaches being tested. The study used a 55k DGT corpus and an 88k public admin corpus for evaluation. A Transformer model with a 16k BPE subword model achieved a BLEU score improvement of 7.8 points compared to a baseline RNN model. The study also shows that Transformer models outperform RNN models in translation accuracy, with the best Transformer model achieving a BLEU score of 60.5 and a TER score of 0.33. The research highlights the importance of subword model selection and hyperparameter optimization in low-resource language translation. The study also addresses the environmental impact of model development, noting that energy consumption during model training was tracked. The results indicate that using a 16k BPE subword model in a Transformer architecture leads to significant performance improvements in low-resource English-Irish translation. The study concludes that hyperparameter optimization and subword model selection are crucial for achieving high performance in low-resource language translation. The research also shows that the Transformer model outperforms Google Translate in translation accuracy for the English-Irish language pair.