Generate requests are sometimes cut short

[mattjhawken]

Two issues have arose when sending generate requests via API. The model output can sometimes be cut short, perhaps just an artifact of the LLMs being used, perhaps not. Likewise sometimes the request gets stuck on the worker (ie. Validator receives request, sends .generate to worker, who does not return anything and we get a 500 internal server error response.