I tested a bit and realized that the models don’t generate proper endings of sentences, as if the response was cut-off at a certain definded character (text length limited in top right menu). Is there a design reason? Is there a way to avoid this?
