This folder contains out logic for generating ParaGuide inferences.
Please note that this logic expects preprocessed data (See data/enron), with (paraphrase, original text) pairs.
To run the full ParaGuide pipeline without preprocessing, see demo/.
example_inference.sh contains an example invocation of the inference script.
inference.py contains our inference logic for both Enron attribute transfer (formality or sentiment) and authorship transfer (with style embeddings).
classifiers.py contains our utilities for loading and using different guidance models.
diffusion_utils.py implements our diffusion inference utilities.
inference_utils.py implements our general inference/configuration utilities.
- Include LUAR guidance support (https://aclanthology.org/2021.emnlp-main.70/)