HSPD detoxifies pretraining corpora via hierarchical semantic-preserving rewriting with Soft Contrastive Decoding, cutting toxicity probability from 0.42 to 0.18 and expected maximum toxicity from 0.43 to 0.20 on GPT2-XL with consistent gains on other models.
Figure 8: Examples of the retrieval results of templated responses for detoxified model
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Detoxification for LLM: From Dataset Itself
HSPD detoxifies pretraining corpora via hierarchical semantic-preserving rewriting with Soft Contrastive Decoding, cutting toxicity probability from 0.42 to 0.18 and expected maximum toxicity from 0.43 to 0.20 on GPT2-XL with consistent gains on other models.