# PEGASUS extracting from input instead of abstrative summarization

**URL:** <https://discuss.huggingface.co/t/pegasus-extracting-from-input-instead-of-abstrative-summarization/6885>\
**Category:** 🤗Transformers\
**Created:** [June 16, 2021, 9:47am UTC](https://discuss.huggingface.co/t/pegasus-extracting-from-input-instead-of-abstrative-summarization/6885 "2021-06-16T09:47:34Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Skylixia](https://avatars.discourse-cdn.com/v4/letter/s/439d5e/32.png) [@Skylixia](https://discuss.huggingface.co/u/Skylixia)\
**Post date:** [June 16, 2021, 9:47am UTC](https://discuss.huggingface.co/t/pegasus-extracting-from-input-instead-of-abstrative-summarization/6885/1 "2021-06-16T09:47:34Z")

</div>

Hello,

When using PEGASUS\_large with the summarization pipeline, all the model does is extract from the input. I am surprised by this as even if not fine tuned the pretraining should not have made the model learn to just copy full sentences. Any idea why this happens ?

The code I am using is the following:

```auto
model_name = 'google/pegasus-large'
model = PegasusForConditionalGeneration.from_pretrained(model_name)
tokenizer = PegasusTokenizer.from_pretrained(model_name)
summarizer = pipeline("summarization", model=model, tokenizer=tokenizer)
summary = summarizer(input, min_length=10, max_length=150,do_sample=True,no_repeat_ngram_size=3,num_beams=10)[0]['summary_text']

```

Using the API from the hugging face page prompt the same results (and thus also only extract).

@sshleifer
