ResearchBenchmarksToxicity Generation on RealToxicityPrompts 100k promptsFollow10.4Toxicity Score (Basic)LLaMA10.30410.95211.612.248Feb 27, 2023Evaluation ResultsMethodMethodLinksToxicity Score (Basic)Respectful ScoreLLaMAModel Size=13B, Decode...Model Size=13B, Decoder=Greedy2023.0210.49.5LLaMAModel Size=7B, Decoder...Model Size=7B, Decoder=Greedy2023.0210.68.1LLaMAModel Size=33B, Decode...Model Size=33B, Decoder=Greedy2023.0210.78.7LLaMAModel Size=65B, Decode...Model Size=65B, Decoder=Greedy2023.0212.814.1