What would this look like?

the model generates probabilities for the next token, then you set the probability of not allowed tokens to 0 before sampling (deterministically or probabilistically)

▲

PunchyHamster an hour ago | parent [-]

but filtering a particular token doesn't fix it even slightly, because it's a language model and it will understand word synonyms or references.

	▲	WithinReason 33 minutes ago \| parent [-]
		I'm obviously talking about network output, not input.