Recently the method for sending out text responses has changed on the bots. Previously, the bots would send multiple shorter messages in response to a prompt, but now the behavior is to send out one longer message instead. This is generally preferable.
However, this has caused a new issue where the text generated runs up against some sort of maximum length and the end of the message is cut off. From the testing that I've done, this limit is around 1000 characters. I'm not sure that is a limit of the LLM itself or because of the character limit of IMs in SL.
Additionally, the bot is now more likely to become unresponsive given a prompt that would require a longer response. I'm guessing that it's generating some sort of error that we can't see, but it never gives a reply and also no longer responds to any further prompts by that particular user until it is restarted manually.
One possible solution to this that retains the new behavior is to break the response up into several messages if it looks like the generated response will be greater than ~800 characters. That will retain the longer potential responses of the previous method while still preventing most multiple-message replies.