There has been a relatively strong local model community since the days of llama-3. IMO, the number of non-technical people interested in qlora, adapters, parameter fine tuning etc is remarkable.

▲

htsh a day ago | parent [-]

that was when RAM and GPUs were cheap. Now that is dying, too.

	▲	cykros a day ago \| parent [-]
		Only people who have no concerns about privacy would give up on it entirely. But yes, it has been a test of just how few tokens you can get by on.