Summary of Parameter-efficient Detoxification with Contrastive Decoding, by Tong Niu et al.
Parameter-Efficient Detoxification with Contrastive Decoding
by Tong Niu, Caiming Xiong, Semih Yavuz, Yingbo Zhou
First submitted to arxiv on: 13 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel controllable text generation technique called Detoxification Generator (DETOXIGEN) is introduced in this paper. DETOXIGEN aims to steer text generation away from undesirable attributes such as toxicity by combining a pre-trained language model with a trained detoxifier. The detoxifier is trained on toxic data and used during generation to produce tokens that the generator should avoid. This approach significantly outperforms previous methods in detoxification metrics while maintaining generation quality. DETOXIGEN also requires minimal additional memory, making it a promising practical solution for real-world applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper makes it possible to control what kind of text is generated by using a new technique called Detoxification Generator (DETOXIGEN). DETOXIGEN helps the generator create text that doesn’t have bad qualities like being mean or rude. It does this by working with a trained “detoxifier” that produces tokens that should be avoided. This makes it much better at removing unwanted styles from generated text. The best part is that it’s really efficient and won’t take up too much memory. |
Keywords
» Artificial intelligence » Language model » Text generation
