Comment by bugglebeetle

Isn’t the tokenization tied to the training of the model?