-
Is there any chance to get support for cyrillc words? When i get an answer in Russian, it feels fine and ok, but amount of tokens rise till 1000 just for 10-20 messages, and console looks like this. So by my feelings each time the new meassage generating, old messages getting in memory and trying to convert russian words into unicode without BOM support (maybe???) so that why amount of tokens rise alot (I dunno, im not a programmer) Also first 10 messages generating a much faster, than a second 10.
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
It is common that non-english text takes up significantly more tokens on most English focused models. You'd have to find a model that was trained specifically for multilingual text. It has nothing to do with the inference software |
Beta Was this translation helpful? Give feedback.
It is common that non-english text takes up significantly more tokens on most English focused models. You'd have to find a model that was trained specifically for multilingual text. It has nothing to do with the inference software