As far as I have read from sentencepiece github, sentencepiece can be done in modes of unigram(default) and bpe and so on. So what exactly do the tokenizers for these mt model is made of ? unigram as default or bpe ?
As far as I have read from sentencepiece github, sentencepiece can be done in modes of unigram(default) and bpe and so on. So what exactly do the tokenizers for these mt model is made of ? unigram as default or bpe ?