LongT5 tGlobal Base Extractive Result


I was looking at the google/long-t5-tglobal-base model (google/long-t5-tglobal-base · Hugging Face) and the texts I tried inputting gave me an extractive summary instead of one that was abstractive (to specify, there did not appear to be any new text generated; each summary was a sentence of the original source text). Since my understanding was that the LongT5 model was for abstractive summarization rather than extractive summarization based on the corresponding research paper, any insight would be greatly appreciated. Thanks in advance!