Erronously Repeating text in API answer

I need help with an issue I have using a python summary function API call to 3.5-turbo-0125 (&1106). I have seen the behaviour over both models:

Infrequently the model produces this weird answer pattern, where it starts out summarizing text as expected and then after, say a page of text, it starts repeating the same few sentences until token limit is hit.

I use a simple function to summarize long text (within token limits!) and maybe it has something to do with the fact that literate over a list of prompts in a for loop to have them all processed sequentially?

def summarize(text, model, prompts):
for prompt in prompts:
#do stuff
return results

But it usually happens in the first iteration.