yxuansu/SimCTG
A contrastive decoding method for improving text generation quality from neural language models.
Not currently ranked — collecting fresh signals.
star history
This repository provides the code and models for a contrastive framework for neural text generation, presented at NeurIPS 2022. The method improves decoding strategies for language models by leveraging contrastive learning principles during generation. It offers an alternative to standard greedy or beam search decoding, addressing issues like repetition and low-quality outputs. The project integrates with Huggingface transformers for easy adoption.
Frequently asked
- What is yxuansu/SimCTG?
- A contrastive decoding method for improving text generation quality from neural language models.
- Is SimCTG open source?
- Yes — yxuansu/SimCTG is open source, released under the MIT license.
- What language is SimCTG written in?
- yxuansu/SimCTG is primarily written in Python.
- How popular is SimCTG?
- yxuansu/SimCTG has 478 stars on GitHub.
- Where can I find SimCTG?
- yxuansu/SimCTG is on GitHub at https://github.com/yxuansu/SimCTG.