Implementation of the MAUVE to evaluate text generation
Project description
MAUVE
This is a library built on PyTorch and HuggingFace Transformers to measure the gap between neural text and human text with the MAUVE measure, introduced in this NeurIPS 2021 paper (Outstanding Paper Award) and this JMLR 2023 paper.
MAUVE is a measure of the gap between neural text and human text. It is computed using the Kullback–Leibler (KL) divergences between the two text distributions in a quantized embedding space of a large language model. MAUVE can identify differences in quality arising from model sizes and decoding algorithms.
Documentation Link GitHub Link
MAUVE is also available via HuggingFace Evaluate!
Features:
- MAUVE with quantization using k-means.
- Adaptive selection of k-means hyperparameters.
- Compute MAUVE using pre-computed GPT-2 features (i.e., terminal hidden state), or featurize raw text using HuggingFace transformers + PyTorch.
- Use with other modalities (e.g. images or audio) by directly passing in pre-computed feature embeddings to our API.
More details can be found in the documentation.
Installation
For a direct install, run this command from your terminal:
pip install mauve-text
Citation
If you find this package useful, or you use it in your research, please cite the following papers:
@article{pillutla-etal:mauve:jmlr2023,
title={{MAUVE Scores for Generative Models: Theory and Practice}},
author={Pillutla, Krishna and Liu, Lang and Thickstun, John and Welleck, Sean and Swayamdipta, Swabha and Zellers, Rowan and Oh, Sewoong and Choi, Yejin and Harchaoui, Zaid},
journal={JMLR},
year={2023}
}
@inproceedings{pillutla-etal:mauve:neurips2021,
title={MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers},
author={Pillutla, Krishna and Swayamdipta, Swabha and Zellers, Rowan and Thickstun, John and Welleck, Sean and Choi, Yejin and Harchaoui, Zaid},
booktitle = {NeurIPS},
year = {2021}
}
@inproceedings{liu-etal:mauve-theory:neurips2021,
title={{Divergence Frontiers for Generative Models: Sample Complexity, Quantization Effects, and Frontier Integrals}},
author={Liu, Lang and Pillutla, Krishna and Welleck, Sean and Oh, Sewoong and Choi, Yejin and Harchaoui, Zaid},
booktitle={NeurIPS},
year={2021}
}
Acknowledgements
This work was supported by NSF DMS-2134012, NSF CCF-2019844, NSF DMS-2023166, the DARPA MCS program through NIWC Pacific (N66001-19-2-4031), the CIFAR "Learning in Machines & Brains" program, a Qualcomm Innovation Fellowship, and faculty research awards.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for mauve_text-0.4.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | ceafe978fd66adf4a2a8bad8c47be417e7306f43a4a4ab9121de01f81fc7e47b |
|
MD5 | d812928b4ac853f0e38496881d1e504d |
|
BLAKE2b-256 | 1f0a7fa7d797479b762e800b57064c5fe861743fe12722292c36de220fc964a2 |