PEI-detector-models

List of deep learning models that are capable of distinguishing computer-generated text from the human-generated text. Each model is inside their respective folder.

Models

• chatgpt-roberta: Recent model (2023) focused on distinguishing chatgpt generated text from a human. They started with the Roberta-base model (which is a masked language model) and then finetuned it in curated data. PAPER CODE.
• openai-roberta-base: Model trained in 2019 for distinguish gpt2 generated text from human. They started with the Roberta-base model (which is a masked language model) and then finetuned it in curated data. PAPER CODE.
• openai-roberta-large: Model trained in 2019 for distinguish gpt2 generated text from human. They started with the Roberta-large model (which is a masked language model) and then finetuned it in curated data. PAPER CODE.

How to run

In each folder there is a run.sh script, just run that. The script will build a virtual environment and install all the dependencies.

Inside chatgpt-roberta-detector folder

$ ./run.sh

Inside openai-reberta-detector folder

This runs the base version

$ ./run-base.sh

This runs the large version

$ ./run-large.sh

Dependencies

The openai-roberta-detector only works with python3.7, so python3.7 must be installed!!! (This is a hard requirement due to the transformer library) The chatgpt-roberta-detector works with the latest version of the required libraries, so it should be good to run.

For both models, I am creating virtual-environments, so it is required to have the package python-venv installed.