List your company

Robustness against harmful content generation on LMSYS-MF attacks

0.81Attack Success RateIT
-0.00120.20940.420.6306Feb 9, 2026

Evaluation results

MethodLinks
ITModel Family=LLaMA 3, Model Size=8B2026.020.81
ITModel Family=LLaMA 3, Model Size=3B2026.020.77
ITModel Family=Gemma 2, Model Size=9B2026.020.75
RLModel Family=Gemma 2, Model Size=9B2026.020.72
ITModel Family=Gemma 2, Model Size=2B2026.020.71
ITModel Family=LLaMA 3, Model Size=1B2026.020.68
RLModel Family=Gemma 2, Model Size=2B2026.020.67
RLModel Family=LLaMA 3, Model Size=3B2026.020.64
RLModel Family=LLaMA 3, Model Size=1B2026.020.61
RLModel Family=LLaMA 3, Model Size=8B2026.020.61
RLBFModel Family=LLaMA 3, Model Size=1B2026.020.07
BSAFE+Model Family=LLaMA 3, Model Size=1B2026.020.06
BSAFE+Model Family=LLaMA 3, Model Size=3B2026.020.05
BSAFE+Model Family=LLaMA 3, Model Size=8B2026.020.05
RLBFModel Family=Gemma 2, Model Size=2B2026.020.05
RLBFModel Family=LLaMA 3, Model Size=3B2026.020.05
BSAFE+Model Family=Gemma 2, Model Size=2B2026.020.04
BSAFE+Model Family=Gemma 2, Model Size=9B2026.020.03
RLBFModel Family=Gemma 2, Model Size=9B2026.020.03
RLBFModel Family=LLaMA 3, Model Size=8B2026.020.03