[Paper link] [Demo link] [Data link]
Work in progress, please revisit this page soon for more in-depth instructions.
Run the commands provided in order to set up your environment:
git clone https://github.com/allenai/real-toxicity-prompts.git
cd real-toxicity-prompts
- Install Anaconda, then run
conda env create -f environment.yml
to create a new Anaconda environment namedrtp
with all necessary dependencies. - Run
conda activate rtp
before running any further scripts or notebooks.
Data processing steps are forthcoming.
We used the following two APIs to run our experiments:
- Perspective API to score text for toxicity
- OpenAI API to generate text with the GPT-3 model
The prompts experiments can be run from scripts/run_prompts_experiments.py
.
Type python -m scripts.run_prompts_experiments --help
to get a list of the available models and arguments that can
be used.