An Efficient Multilingual Language Model Compression through Vocabulary Trimming