The paper introduces TOWER, an open multilingual large language model (LLM) designed for translation-related tasks. The authors propose a method to tailor LLMs for multiple translation workflows by extending the multilingual capabilities of LLaMA-2 through continued pretraining on a mixture of monolingual and parallel data, creating TOWERBASE. They then fine-tune this model on a dataset of high-quality and diverse instructions for translation-related tasks, resulting in TOWERINSTRUCT. TOWERINSTRUCT consistently outperforms open alternatives on various translation tasks and is competitive with closed-source models like GPT-4. The paper also includes a detailed evaluation of the model's performance on translation quality, automatic post-edition, named entity recognition, and grammatical error correction. The authors release the TOWER models, the specialization dataset TOWERBLOCKS, an evaluation framework TOWEREVAL, and a collection of model generations to facilitate future research.The paper introduces TOWER, an open multilingual large language model (LLM) designed for translation-related tasks. The authors propose a method to tailor LLMs for multiple translation workflows by extending the multilingual capabilities of LLaMA-2 through continued pretraining on a mixture of monolingual and parallel data, creating TOWERBASE. They then fine-tune this model on a dataset of high-quality and diverse instructions for translation-related tasks, resulting in TOWERINSTRUCT. TOWERINSTRUCT consistently outperforms open alternatives on various translation tasks and is competitive with closed-source models like GPT-4. The paper also includes a detailed evaluation of the model's performance on translation quality, automatic post-edition, named entity recognition, and grammatical error correction. The authors release the TOWER models, the specialization dataset TOWERBLOCKS, an evaluation framework TOWEREVAL, and a collection of model generations to facilitate future research.