BERT, a transformer-based model for natural language processing, requires a tokenizer to convert text into integer tokens. This article outlines the process of training a WordPiece tokenizer in line with BERT's original specifications. The author explains that while BERT is efficient enough to run on a personal computer, the right dataset is essential for effective training. The article is structured into two main sections: selecting a dataset and training the tokenizer. The dataset's size and characteristics heavily influence the tokenizer's performance. The author provides practical guidance on how to choose an appropriate dataset, emphasizing the importance of diversity and representation in the text data for optimal results. In the training section, the article details the steps to create and fine-tune the tokenizer. It goes through the technical aspects, offering insights into parameters and configurations that affect the tokenizer's quality. Specific examples illustrate how to implement these steps using programming libraries. This practical approach helps readers understand the underlying processes, making it easier to replicate the training of a WordPiece tokenizer for BERT models.