Abstarct: Persian is a low-resource language and thus abstractive-summarization is a considerable challenge with the lack of an efficient dataset. We gathered a dataset suitable for fine-tuning a pretrained transformer model, then employed the mt5 pre-trained model to train an abstractive tixtle-generation model on our introduced dataset. The model achieved the Rouge-1 score of 46.32 with coherent outputs. The results are competitive with state-of-the-art similar English works.