package for CTRLsum (easy to use)

[hyunwoongko]

Thanks for cool research. The results of this research were quite appealing to me, so I made the package to make it easier to use. The package is called summarizers. I ported the model you provided to Hugging Face Transformers, and I made it for use with codes 2 to 3 lines. If anyone wants to try this model more easily, it would be good to use.

https://github.com/hyunwoongko/summarizers

Thanks for great research again.

[jxhe]

Thank you very much! I added references to your repo as well as the huggingface interface into README.

[hyunwoongko]

I ported the following name on the Hugging Face.

• https://huggingface.co/hyunwoongko/ctrlsum-cnndm
• https://huggingface.co/hyunwoongko/ctrlsum-arxiv
• https://huggingface.co/hyunwoongko/ctrlsum-bigpatent

Thanks a lot!

[jxhe]

Thank you! I have a question for the unconditional summarization example: the unconditional summaries are generated with empty keywords/prompts in the package, right? I am asking just for clarification since it is different from the paper which uses automatic keywords, while I think it is still good and simple without an additional tagger.

[hyunwoongko]

Yeah. because It is just first draft version. I will add BERT tagger later (next version) Thanks :)

[jxhe]

Hi, the pretrained model checkpoints are already avaiable at huggingface, but we don't have a training script to train CTRLsum using Huggingface.

However, you can train CTRLsum with huggingface easily -- just run our preprocessing script to prepare the data, then run huggingface's summarization example script here following our training hyperparameters.