The paper introduces TASTe (Translate through Self-Reflection), a novel framework designed to enhance the translation capabilities of large language models (LLMs). TASTe employs a two-stage inference process where LLMs first generate preliminary translations and simultaneously assess their quality. In the second stage, these preliminary translations are refined based on the quality predictions. This self-reflection process is achieved through multitask fine-tuning on a dataset that includes Quality Prediction, Basic Translation, and Draft Refinement tasks. The effectiveness of TASTe is demonstrated through experiments on the WMT22 benchmark, showing superior translation quality compared to existing methods. The approach leverages LLMs' ability to predict translation quality, refining initial drafts to produce more accurate and diverse translations. The paper also discusses the limitations and future directions, highlighting the need for further exploration of multilingual knowledge and computational efficiency.The paper introduces TASTe (Translate through Self-Reflection), a novel framework designed to enhance the translation capabilities of large language models (LLMs). TASTe employs a two-stage inference process where LLMs first generate preliminary translations and simultaneously assess their quality. In the second stage, these preliminary translations are refined based on the quality predictions. This self-reflection process is achieved through multitask fine-tuning on a dataset that includes Quality Prediction, Basic Translation, and Draft Refinement tasks. The effectiveness of TASTe is demonstrated through experiments on the WMT22 benchmark, showing superior translation quality compared to existing methods. The approach leverages LLMs' ability to predict translation quality, refining initial drafts to produce more accurate and diverse translations. The paper also discusses the limitations and future directions, highlighting the need for further exploration of multilingual knowledge and computational efficiency.