Similar to this post, I was wondering if there’s a way to adjust abstractive summaries to give a more tailored output via HF.
For example, with T5 or GPT2 is there a way to either (a) set a temperature or (b) prioritize specific words/sentences more heavily?
One of the issues I am facing is that the abstractive summary for most news articles from the WikiGold dataset are simply extractive rather than abstractive.