Polyjuice

🚀 Polyjuice

Polyjuice is a general - purpose counterfactual generator. It can generate counterfactual texts, which is useful for explaining, evaluating, and improving models.

🚀 Quick Start

✨ Features

• It is a ported version of the general - purpose counterfactual generator Polyjuice.
• For more code release, you can refer to this github page.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer, AutoModelWithLMHead
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM

model_path = "uw-hai/polyjuice"
generator = pipeline("text-generation", 
    model=AutoModelForCausalLM.from_pretrained(model_path), 
    tokenizer=AutoTokenizer.from_pretrained(model_path),
    framework="pt", device=0 if is_cuda else -1)

prompt_text = "A dog is embraced by the woman. <|perturb|> [negation] A dog is [BLANK] the woman."
generator(prompt_text, num_beams=3, num_return_sequences=3)

📚 Documentation

BibTeX entry and citation info

@inproceedings{polyjuice:acl21,
    title = "{P}olyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models",
    author = "Tongshuang Wu and Marco Tulio Ribeiro and Jeffrey Heer and Daniel S. Weld",
    booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics",
    year = "2021",
    publisher = "Association for Computational Linguistics"