That is fair stance to take, and understandable as at the moment as especially with American models the token rate for even trivial decompliation is quite silly, but like image and video generation that’ll get reduced drastically soon and you’ll be able to run it on a 2090 in reasonable timespace.
That is fair stance to take, and understandable as at the moment as especially with American models the token rate for even trivial decompliation is quite silly, but like image and video generation that’ll get reduced drastically soon and you’ll be able to run it on a 2090 in reasonable timespace.