---
language: en
---

# CTRL44 Simplification model

This is a pretrained version of the controllable simplification model presented in the NAACL 2022 paper "Controllable Sentence Simplification via Operation Classification". It was trained on the IRSD simplification dataset.

A control token is expected at the start of input sequences to dictate which simplification operation should be performed. This can either be done manually or with an operation classifier like [this one](https://huggingface.co/liamcripwell/ctrl44-clf).

Possible control tokens are: "\<ident\>", "\<para\>", "\<ssplit\>", and "\<dsplit\>".

## How to use

Here is how to use this model in PyTorch:

```python
from transformers import BartForConditionalGeneration, AutoTokenizer

model = BartForConditionalGeneration.from_pretrained("liamcripwell/ctrl44-simp")
tokenizer = AutoTokenizer.from_pretrained("liamcripwell/ctrl44-simp")

text = "<para> Barack Hussein Obama II is an American politician who served as the 44th president of the United States from 2009 to 2017."
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, num_beams=10, max_length=128)
```