Increase speed Huggingface tokenizer ouput

Question

I need to get the last layer of embeddings from a BERT model using HuggingFace. The following code works, however is extremely slow, how can I incre