Multiple times deepseek r1 stops generating tokens in the middle of thinking/responding when reaching around 5800 tokens (not sure that number plays a role or it just happened). Sometimes this is even in the middle of thinking after the very first prompt. Are others experiencing this? Not sure if it is sth about kagi or the api provider. Obviously it is hard to use a reasoning model like that.
The model should not stop the token generation. If it stops for some reason, it would be nice to be possible at least to contunue from where it stopped, rather having to regenerate answer from the beginning.
