Yes. I split the text into sentence and append sentences to a chunk until the max context window is reached. The context window size is dynamic for each article so that each chunk is roughly the same size. Then I just do a mean pool of the chunks for each article.