# Datasets for generating longer summaries

**URL:** <https://discuss.huggingface.co/t/datasets-for-generating-longer-summaries/2479>\
**Category:** Models\
**Created:** [December 3, 2020, 3:36pm UTC](https://discuss.huggingface.co/t/datasets-for-generating-longer-summaries/2479 "2020-12-03T15:36:43Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Abhijith](https://avatars.discourse-cdn.com/v4/letter/a/82dd89/32.png) [@Abhijith](https://discuss.huggingface.co/u/Abhijith)\
**Post date:** [December 3, 2020, 3:36pm UTC](https://discuss.huggingface.co/t/datasets-for-generating-longer-summaries/2479/1 "2020-12-03T15:36:43Z")

</div>

Hi,  
I was looking into generating summaries for big articles (3-4k tokens) using Longformer (summary length 200-300 token). I tried using [this](https://huggingface.co/patrickvonplaten/longformer2roberta-cnn_dailymail-fp16) model by @patrickvonplaten . Since it was trained CNN/Daily mail the summaries are pretty small (3-4 lines). I was going to fine-tune to generate longer summaries. Are there any such datasets that have bigger summaries? I took a look at datasets listed [here](http://nlpprogress.com/english/summarization.html) but could not find anything.

If there aren’t any datasets like this what are some other strategies that I can use? Filter out the shorter summaries and only use the big ones?
