Will the model give good results if it is trained to generate an abstractive summary of a large piece of text? Will the model give good results if it is trained in generating an abstractive summary of 3-4 sentences? If not what changes would you suggest for good results?