[zaptrem]

This is quite similar to the models used by all the AI music generators. Some feed the tokens into a language model to generate music, some replace the tokenization part with an alternative that gives a continuous representation for diffusion models.