[anuramat]

wdym by "prompt and vector is small"? small as in "less tokens"? that should be a positive thing for any kind of estimation

in any case, how is this specific to transformers?