Does it support paragraph prediction?

[LucasAndLevinAndJadon]

I am predicting paragraph based on this model.
The output separates sentences randomly.
Certainly it has something to do with the fast the my data text is not clean.
I wonder whether this is a need for some others too.
Thank you.

[gench]

BERT has a max length limit of 512 tokens. It does not do any sentence-segmentation but treat the given input text as it is. You can segment your paragraph into sentences and calculate the sentence-based sentiment. Then, calculate paragraph-level sentiment over those.