Paper Title Generator
This repository contains source code for predicting computer science paper titles from abstracts using a BERT2BERT transformer model.
A demo of the inference web interface can be found at https://paper-titles.ey.r.appspot.com/.
A pre-trained model is available on HuggingFace: Callidior/bert2bert-base-arxiv-titlegen
Contents
The repository comprises the following directories:
training
: Code for training the BERT2BERT model on the arXiv dataset.self-hosted
: Code for the web interface performing inference locally.hosted-inference
: Code for the web interface using the HuggingFace hosted inference API for inference.